Assessing the Role of Safety Perception in the Relationship Between Street Environments and Women’s Stay Behavior, Using Multi-Source Big Data

Abstract

This study establishes an impact pathway hypothesis connecting street environments, safety perception, and women’s stay behaviors based on spatial cognition theory and the lens model theory. It examines the driving factors behind female environmental perception and behavioral patterns in urban streetscapes by integrating street view imagery and facility point data. Random forest models and questionnaire surveys were employed to evaluate the sense of security, and structural equation modeling was used to quantify environmental features, safety perception, and behavioral indicators. The results demonstrate that four street environment dimensions, functionality formats, interface morphology, spatial quality, and street facilities, exert varying degrees of positive or negative influences on women’s safety perception and behaviors. Perceived safety serves as a significant mediator in the environment–behavior pathway, with functionalities, spatial quality, and interface morphology exhibiting sequentially decreasing effect magnitudes in this mediated relationship, while street facilities indirectly affect staying behaviors exclusively through the safety perception mechanism.

1. Introduction

The New Urban Agenda, adopted at the United Nations Conference on Housing and Sustainable Urban Development (Habitat III) in 2016, advocates governments worldwide to establish gender-equal, just, safe, and healthy sustainable cities and human settlements [1]. An increasing number of cities globally have begun integrating a gender perspective into urban planning and governance systems. With the advent of postmodern society, the shift in cultural and spatial scales led to the emergence of feminist geography as a distinct field within human geography. Focusing on space and gender as its core subjects of inquiry, it examines how gender and geography mutually constitute and transform one another, as well as how gender differentiation permeates social life. Feminist geography, as a discipline that initially explored the relationship between women and space, has systematically described and deconstructed the challenges women face in urban environments [2]. Its critical research on material spaces mainly focuses on how spatial structures influence women’s opportunities for survival and development equity through multiple pathways. Influenced by feminist geography, urban planning and architecture have undergone a cognitive shift, establishing an intrinsic connection between spatial justice and gender equality. It is now widely recognized that minimizing inequalities in spatial use and experience is a fundamental pathway toward achieving gender equality in space [3]. Current research on women in urban planning has evolved from a singular female perspective to a more inclusive gender-sensitive approach [4], covering women’s health [5,6,7,8], travel behavior [9,10,11], spatial perception [12,13,14], and equity in development opportunities [15]. The core of urban research from a female perspective focuses on women’s safety in public spaces and gender-equitable practices in urban planning [3].

Safety is a fundamental need in urban perception, and it is also one of the core demands of women for the built environment. A positive sense of security will promote urban activities. In urban spaces, individual stay behaviors primarily include waiting, resting, and other activities characterized by relative spatial stillness. In city streets, individual stay behavior is closely linked to spatial vitality [16]. However, individual behavioral decisions are not only shaped by the objective urban environment but also influenced by subjective perceptions [17]. Spatial cognition theory posits that perception is a critical link connecting the objective environment with behavioral responses [12,18]. Individuals receive information about the urban environment through sensory organs and process it based on prior knowledge and experience, forming a subjective perception of their surroundings. This perception not only directly influences individual behavioral reactions but also provides a basis for behavioral decision making. In empirical studies on environment–behavior relationships, considerable attention has been given to the direct relationships between paired variables such as “environment–perception” and “environment–behavior,” while relatively less emphasis has been placed on the underlying mechanisms. The influence of the environment on behavioral decisions often does not occur directly; rather, it is mediated through perceptual processes. Therefore, it is essential to systematically examine the complete pathway of “environment—perception—behavior.”

Focusing on the sense of security among women, existing research has confirmed that the urban built environment significantly influences female safety [12,19,20,21,22], while these environmental factors also affect women’s daily behaviors and spatial decision making through both positive and negative effects [10,11,23]. However, there remains a notable gap in current studies regarding the role of safety perception in the pathway of “built environment–women’s daily behavior.” In the field of urban planning, when examining the impact of the built environment on residents’ behavior, there has been a tendency to adopt a generalized and averaged perspective, often overlooking the specific needs of women. This approach may result in urban designs that fail to address the deeper requirements of diverse groups, ultimately affecting their well-being, equity, and the overall vitality of the city. Incorporating safety perception, as a key variable in the study of how street environments influence female stay behavior, not only helps reveal the complex pathway of “built environment–perception–behavior” but also provides a theoretical bridge for understanding the intrinsic connection between urban physical spaces and women’s behavioral decision making. This, in turn, promotes urban spatial justice and advances gender equality.

This study constructs a street environment indicator system, a female safety perception evaluation model, and a female stay behavior indicator system based on facility point data (point of interest, POI; such as shops, hotels, and restaurants) and street view images. Utilizing deep learning and on-site observations to collect data for each indicator, the research employs structural equation modeling (SEM) to establish relationships among the three components. It also examines the role of safety perception in the pathway through which street environments influence women’s stay behavior.

The rest of this article is organized as follows: after a brief review and theoretical hypotheses in Section 2, we describe the research method in Section 3. In Section 4, we introduce the study area and data sources, the data analysis results are in Section 5, and Section 6 is the results discussion. Section 7 concludes the study and addresses its limitations.

2. Literature Review and Theoretical Hypotheses

2.1. Spatial Cognition Hypothesis and Lens Model Theory

Human spatial behavior is determined by how individuals interpret, feel, and perceive the external world, as spatial perception serves as the foundation for behavioral decision making [24]. The spatial cognition hypothesis suggests [25] that individuals develop multi-dimensional spatial perceptions under the influence of the physical environment, while these perceptions also mediate and moderate the relationship between the material environment and individual behavior [26]. Individuals store information about the built environment in their cognitive maps, and their ability to process and retain information is largely shaped by personal attitudes and perceptual preferences. These factors further inform decision making, ultimately influencing individual behavior. Within the lens model theory (Figure 1), the relationship between the built environment and spatial behavior is decomposed into five dimensions: physical attributes, perceptual/cognitive judgments of physical attributes, affect, well-being, and spatial behavior [26]. In the framework, each component has a probability (P) of influencing the next level. Among these, physical attributes (A) represent measurable indicator elements derived from the objective urban environment (e.g., vegetation density, spatial dimensions). Perception/cognition (J) refers to an individual’s multi-dimensional perception of the space (e.g., aesthetic appeal, sense of safety) [27]. The theory posits that human perception is an active process of filtering useful imagery from the environment. While urban environments provide numerous elemental indicators, only a subset of these elements actually influence the perceptual process. After the brain receives environmental stimuli and generates perceptual outcomes, these results subsequently influence behavioral responses.

Based on the theoretical content above, it is evident that human behavioral decision making in space is not a direct reaction to the physical environment but is instead built upon perception and cognition. Spatial cognitive theory reveals that there exists a critical and measurable mediating factor between the objective environment and individual behavior—namely, individual perception. The lens model provides a well-established analytical framework to quantify this mediating process. Thus, the relationship between the built environment and individual behavior can be deconstructed into a causal chain: urban spatial environment—perception—behavior. More specifically, human perception filters “useful indicators” from the complex urban environment. Although the environment offers a vast array of elements, only a subset of key features is captured and utilized by individuals for judgment, subsequently influencing their behavioral decisions. The purpose of establishing the “urban spatial environment—perception—behavior” framework is to uncover the process of individual behavioral decision making and to construct an interpretable and measurable theoretical bridge between the objective environment and subjective behavior. This, in turn, provides a precise classification basis and design strategies for optimizing the urban built environment.

2.2. Impact of the Built Environment on Stay Behavior

Stay behavior refers to pedestrians’ relatively immobile activities in urban spaces, such as waiting, socializing, or resting. This behavior is self-determined, as individuals have sufficient conditions to flexibly choose their staying locations and activity spaces. Jan Gehl’s Copenhagen studies revealed that stay behavior correlates strongly with urban vitality, and prolonged stays depend on both the stayability of environmental settings and people’s spatial perception intentionality [28]. In different types of urban spaces, existing studies have confirmed that built environment elements exert varying degrees of influence on stay behaviors. In traditional village streets, street facility configuration and interface permeability are key factors affecting the duration of stay [29]. In underground commercial streets, the scale of pedestrian space has a significant positive correlation with stay behavior [30]. In studies on urban streets, some scholars have used stay behavior ratio and density as indicators to quantify stay behavior [16]. Factors such as street spatial scale, functionality, and leisure facilities exert differentiated influences on stay behavior [30]. In research on urban river corridors, stay behavior has been adopted as one of the indicators to quantify pedestrian movement trajectories. Visual environment elements, such as plant quantity and facility density, were found to affect both staying duration and staying density [31]. In studies on the relationship between the built environment and stay behavior, stay behavior is often used as a proxy indicator to quantify pedestrian activity, with a predominant focus on duration of stay. However, there has been relatively little discussion on the nature of stay behavior itself. And some of the behavioral information embedded in staying activities has been overlooked.

2.3. Impact of the Built Environment on Female Safety

Sense of security refers to the anticipation of potential physical or psychological danger or risk, coupled with an individual’s feeling of capability or incapability in coping with such situations [32]. It is defined as a subjective feeling and belief of being protected from harm, threat, or danger in one’s environment.

For women, a sense of security is a prerequisite for engaging in various behavioral activities in urban spaces. Relevant studies indicate that women generally perceive lower levels of safety in urban environments compared to men [33]. Environmental psychology divides spatial perception into four stages: sensation—perception—cognition—behavior [25,34]. Specifically:

(1) Sensation: The process by which sensory organs receive environmental stimuli. This study focuses on visual signal reception. (2) Perception: Building upon sensation, this stage integrates prior knowledge and experience to form a comprehensive interpretation of the environment. (3) Cognition: Based on perception, this stage involves emotional processing and logical reasoning, incorporating individual cognitive abilities and contextual factors. (4) Behavior: The stage where individuals store perceived environmental information and execute actionable responses.

Based on the aforementioned perceptual process, this study analyzes women’s psychological safety from a visual perspective. Female individuals receive stimuli from the urban built environment through visual channels, which are then filtered through personal knowledge and gendered experiences, resulting in emotional responses to urban settings that ultimately influence behavior. From the perspective of built environment safety, research based on crime prevention through environmental design (CPTED) theory [35,36] indicates that objective physical elements of the built environment, such as untidy streetscapes, vacant lots, and disordered street furniture or signage, can trigger women’s perception of spatial disorder and fear of crime, thereby reducing their sense of security. From the perspective of behavioral activity safety, studies informed by the “eyes on the street” concept, routine activity theory, and related scholarship [37] demonstrate that factors like appropriate vehicular activity, vibrant street-level commerce, and well-maintained road conditions significantly enhance women’s perceived safety and behavior. In contrast to visually well-regulated and officially monitored spaces, women exhibit heightened fear in chaotic, disordered, or isolated environments, often altering their routes and activity patterns in response. Therefore, integrating this understanding of female perception with existing literature, we propose a two-dimensional framework of women’s sense of security from an environmental stimulus perspective: built environment safety, behavioral activity safety.

As essential users of urban spaces, women exhibit significant differences from men in terms of environmental perception, spatial experience, and activity patterns. Women generally perceive higher safety risks in society—a perception influenced not only by objective crime rates but also by subjective psychological factors, details of the built environment, and socio-cultural influences. Ouali et al. [38,39] found that women are 10% more likely than men to perceive urban environments as unsafe, which directly affects their travel choices. For a long time, urban planning and design have often adopted a gender-neutral perspective, potentially overlooking the heterogeneity between genders in terms of physiology, social roles, and safety needs. This oversight may inadvertently exacerbate marginalized spatial experiences for certain groups [39]. When urban spaces fail to adequately address women’s safety needs, they may respond by shortening their stay, avoiding certain areas or times, or altering their behavior to enhance their sense of security. Such adaptations can restrict women’s behavioral freedom and limit their social participation and access to urban resources, ultimately impairing both individual well-being and the vibrancy and diversity of urban public spaces.

Research on women’s safety in urban spaces can be divided into two directions: studies on the development of women’s safety issues and research on spatial security perception. Scholars have examined various built environments, such as urban public spaces [40], underground public spaces [4], urban village alleys [41], and streets [21], from a female perspective, confirming that built environment elements significantly influence women’s sense of security.

Methodologically, traditional approaches primarily rely on interviews, questionnaires, and safety audit tools to assess women’s safety perceptions. Some researchers have used these methods to conduct in-depth interviews and empirical studies on women’s security perceptions in urban stream corridors and urban residential areas [19,21]. However, such methods demand substantial human and material resources and struggle to collect large-scale data for extensive regional safety perception evaluations. The emergence of street view imagery combined with deep learning techniques has overcome this limitation. Previous studies have employed fully convolutional networks (FCNs) and random forest algorithms (RF) alongside street view images to conduct large-scale urban safety perception assessments [21,34], providing new possibilities for further research on urban security perception.

From a methodological perspective, quantitative approaches dominated by machine learning employ random forest ensembles to model the relationship between street environment data and individual quantitative safety ratings, establishing a predictive framework. This model identifies mathematical patterns linking environmental features and subjective scores, thereby generating evaluated safety perceptions. The underlying individual rating data must be quantized into graded levels to comply with computational requirements of mathematical models. While this method enables large-scale urban perception assessment, it may oversimplify the emotional and experiential nuances of individual perceptions in urban spaces due to the quantization of safety levels, thus presenting certain limitations. On the other hand, qualitative methods such as questionnaires and interviews offer deeper insights into individual perceptual experiences and emotional responses. However, their application in large-scale urban perception studies is constrained by high demands on time, labor, and resources. To address these limitations, Hu et al. [4], in their study on safety perception in residential public spaces, utilized safety audit tools and KH Coder for textual analysis to extract perceptual outcomes from interviews Similarly, Qin et al. [42] adopted a mixed-method approach in research on urban environmental safety, combining qualitative and quantitative techniques, weighted integration of questionnaire responses [43], and machine learning outputs, to derive comprehensive safety perception results.

In conclusion, to achieve a more holistic evaluation of sense of security, this study adopted a hybrid methodology, integrating questionnaire surveys with machine learning techniques to obtain robust and nuanced safety assessment outcomes.

2.4. The Role of Perception in the Path of Built Environment Influence on Individual Behavior

In spatial cognition hypothesis, individual behavioral decisions are influenced by the understanding and perception of the external world environment, with spatial perception serving as the foundation for decision making and behavior [23]. Individual spatial perception needs can generally be categorized into basic levels such as a sense of security and comfort, intermediate levels like a sense of interest and mystery, and higher levels including a sense of belonging and identity [25]. Previous studies have explored the influence pathway of “built environment–perception–behavior,” covering aspects such as restorative perception in urban green landscapes and its impact on public health behaviors [44], the perception of street landscape environmental quality and its effect on individual environmental preferences [45], differences in built environment perception in existing residential areas and their influence on the health behaviors of the elderly [46], as well as the impact on physical activity intentions [47]. These studies have validated that urban environmental elements can affect individual perception levels and subsequently influence behavior.

In the field of urban safety perception, existing studies have explored the relationship between sense of security and individual behavior. Qu et al. [48], in their research on the perception of the urban built environment and women’s cycling behavior, found that a sense of security positively influences the probability of cycling among women. Battiston et al. [49,50], while analyzing gender inequality in urban cycling, identified a positive correlation between female cycling rates and urban road safety. Van Cauwenberg et al. [51] demonstrated in a study on older adults’ walking behavior that insecurity negatively affects walking behavior among the elderly. Lee and Park [52] conducted a mechanistic study on neighborhood environmental safety and walking behavior among older Korean women. Wu Wanshu et al. [53], using two commercial mixed-use streets in Xiamen as case studies and employing indicators such as safety perception, street environment, and stationary activities, applied a partial least squares path model to investigate the underlying mechanisms among these factors. This study confirmed the mediating role of safety between the objective urban environment and stationary activities.

In studies on urban sense of security and individual behavior, research questions have predominantly focused on the direct relationship between perceived environmental safety and individual behavior. These studies often treat safety either as an outcome variable of the built environment or as a direct influencing factor of individual behavior. However, few have considered the mediating role of sense of security, and even fewer have incorporated the built environment, sense of security, and individual behavior into a unified research framework. As a result, there is a lack of systematic analysis regarding the pathways of influence among these three variables.

2.5. Research Hypotheses

Based on the spatial cognition hypothesis, the lens model theory, and literature review, this study constructs a hypothesized pathway regarding the role of safety perception in the influence of street environments on women’s stay behavior. Firstly, it is hypothesized that women’s stay behavior is influenced by urban street spatial environmental elements (Hypothesis Path H3). Secondly, street environmental elements are also considered external direct factors shaping women’s safety (Hypothesis Path H1). Finally, women’s safety in street spaces serves as a mediating variable affecting the pathway from street spatial environment to stationary behavior (Hypothesis Path H2), with the decision-making performance in stay behavior representing the outcome of this pathway.

Accordingly, the study proposes the following hypotheses: (1) How do urban street environmental elements influence female safety and their stay behavior? (2) Does safety perception play a mediating role in the impact of street environmental elements on women’s stay behavior? (3) What are the differences and characteristics in the mediating effects of various street environmental elements on women’s sense of safety?

3. Research Method

3.1. Research Framework

The research framework of this study is divided into four main stages (Figure 2): firstly, constructing a street environmental element indicator system as the independent variable. Utilizing point of interest (POI, e.g., shops, hotels, restaurants) data and street view imagery, semantic segmentation processing is applied to obtain the proportional composition of street elements—specifically, the ratio of pixels occupied by each environmental element to the total pixels in the image. Subsequently, the indicator values are calculated based on the formulas defined in the street indicator system. Secondly, establishing a safety perception evaluation model from a female perspective, and using questionnaire surveys to obtain safety perception assessment results as the mediating variable. Based on street view images and the proportion of street elements coupled with safety ratings from female volunteers, a random forest model is employed to develop a female safety perception evaluation model, yielding safety perception assessment data. Thirdly, developing a female stay behavior indicator system as the dependent variable. Behavioral data is obtained through street point observations, and subsequent calculations generate the indicator data for stay behavior. Finally, constructing a structural equation model (SEM). Using the independent variable, mediating variable, and dependent variable data, a structural equation model is built to examine the mediating effect of safety perception in the pathway between street environment and women’s stay behavior.

3.2. Construction of Street Environment Element Indicator

To ensure the objectivity and comprehensiveness of street environmental element selection and facilitate a better discussion on the impact of street environments on women’s safety perception and stay behavior, we reviewed recent studies related to street spaces [54,55,56]. Previous research indicates that the distribution of functional facilities and the variation in interface morphology along streets significantly influence individuals’ sense of safety and behavior [17,57]. Elements such as greenery and sky visibility in street spaces may positively affect individuals’ safety [41], while the allocation of street infrastructure has a certain degree of influence on individuals’ stay behavior [32]. Based on the indicator systems from relevant studies, street spatial environments were categorized into four dimensions: functionality, interface morphology, spatial quality, and street facilities (

Table 1. Street Environment Indicator System.

Primary Indicators	Secondary Indicators	Formula	Quantitative Explanation
Functionality	Functional Density (F1)	PD = NPOI/L	NPOI refers to the total number of POIs within the buffer zones on both sides of the street, L represents the street length; this metric reflects the density of various facilities along the street.
	Functional Mix (F2)	$\mathrm{X}={\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}}{(\mathrm{P}}_{\mathrm{i}}\times \ln {\mathrm{P}}_{\mathrm{i}})$	n represents the number of POI categories, and Pi denotes the proportion of the i category of POI among all POIs on the street; this reflects the functional diversity (or mixed-use degree) of street facilities.
	Pedestrianization Degree (F3)	W = Si + Pi + BPi + SLi	Si, Pi, BPi, and SLi represent the pixel counts of sidewalks, pedestrians, bicycles, and traffic signals, respectively, reflecting the street’s support level for pedestrian behavior.
	Motorization (F4)	M = Ri + Ci + Mi + Bi + TUi + Ti + SLi	Ri, Ci, Mi, Bi, TUi, Ti, SLi represent the pixel counts of motor lanes, cars, motorcycles, buses, trucks, trailers, and traffic signals, respectively, reflecting the street’s capacity and support level for motorized traffic.
	Visual Walkability (F5)	WV = Si/(Si + Ci + TUi + Bi + Ri)	Si represents the pixel count of sidewalks, while Ci, TUi, Bi, and Ri denote the pixel quantities of cars, trucks, buses, and roadways, respectively. These parameters collectively reflect the street’s visual walkability and pedestrian-friendliness.
Interface Morphology	Interface Complexity (IM1)	RC = Nsy/L	Nsy represents the number of visible street signs along the street segment, while L denotes the street length. These parameters collectively reflect the diversity of street signage.
	Sidewalk Area Ratio (IM2)	SA = SS/SR	SS denotes the sidewalk area of the street segment, and SR represents the total area of the street segment. These parameters collectively reflect the walkability interface level of the street segment.
	Sidewalk Height Difference (IM3)	D = Ns * 0.15	Ns represents the number of steps, where the elevation difference of the sidewalk equals the number of steps multiplied by 0.15 m (Ns × 0.15 m). This parameter reflects the street’s elevation variation.
	Interface Transparency (IM4)	IP = Gi/(Bi + Wi)	Gi represents the pixel count of transparent interfaces along the street segment, while Bi and Wi denote the pixel quantities of buildings and walls, respectively. These parameters collectively reflect the permeability level of street interfaces.
	Scenario Diversity (IM5)	Ri = d	Ri quantifies the richness of streetscape elements in the i-th image, while d represents the count of distinct streetscape element types per image. These metrics collectively characterize the relative diversity of street elements.
Spatial Quality	Green View Index (SQ1)	GV = Gi/A	G represents the pixel count of trees in the streetscape image, while A denotes the total pixel count of the image. These parameters collectively reflect the street greening condition.
	Sky View Factor (SQ2)	Oi = Si/A	Oi represents the proportion of sky pixels to total pixels in the i-th image, where Si denotes the sky pixel count and A indicates the total image pixels. This metric reflects the visible sky proportion in human perspective, influencing both visual openness of the streetscape and perception of natural light availability.
	Enclosure (SQ3)	ED = (Bi + Wi + Gi)/A	Bi, Wi, and Gi represent the pixel counts of buildings, walls, and trees in the streetscape image, respectively, while A denotes the total image pixels, collectively reflecting the degree of street enclosure.
	Spatial Congestion (SQ4)	Vi = Pi + Bki	Vi is the total number of pedestrians and bicycles in the i-th image; Pi represents the number of pedestrians; Bki denotes the number of bicycles, reflecting the crowding level of the street area.
	Street Height-to-Width Ratio (SQ5)	P = Lb/H	Lb represents the proportion of buildings within the street, and H denotes the average of the road and sidewalk proportions; reflecting the spatial compactness.
Street Facilities	Security Facilities Ratio (SF1)	S = Ps/Pt * 100%	S represents the percentage of safety facilities in the image; Ps is the total number of pixels identified by the model as sidewalk elements; Pt denotes the total recognized pixels in the image. Safety facilities are defined as the combined percentage of (surveillance cameras + traffic signs + streetlights + notice boards), reflecting the distribution of safety infrastructure in the street.
	Traffic Signs Ratio (SF2)	IT = T/R	T represents the total pixel count of traffic signals and road signs in the street view image, while R denotes the combined pixel area of vehicle lanes and pedestrian walkways. This metric reflects the distribution of traffic signage across the street.

), The calculation methods for each environmental indicator are shown in the table.

3.3. Construction of Female Safety Evaluation Model

Based on the literature review, and from the perspective of visual reception of environmental stimuli, women’s sense of security is categorized into two dimensions: built environment safety and behavioral activity safety (

Table 2. Safety Perception Indicator System.

Primary Indicators	Secondary Indicators	Measurement Method	Indicator Explanation
Female Safety	Built Environment Safety	Questionnaire Survey and Evaluation Model	The sense of safety derived from elements of the physical street space, such as trees, buildings, etc.
	Behavioral Activity Safety	Questionnaire Survey and Evaluation Model	The sense of safety arising from behavioral dynamics in street spaces, such as vehicular traffic volume, commercial density, and overall activity levels.

). To obtain comprehensive and objective safety evaluation results, this study adopts a hybrid methodology combining questionnaire surveys and machine learning techniques.

In the questionnaire survey, street view images were used to collect safety perception evaluations through offline interviews. The survey questionnaire incorporates a psychological safety scale and is divided into two main sections: Perceived Safety of the Built Environment: participants were sequentially asked about the impact of interface morphology, spatial quality, and street facilities on their sense of safety at different street locations. Behavioral Activity Safety: participants were sequentially questioned about functionalities on their sense of safety at the same street locations.

In terms of machine learning methods, the study used the DeepLabv3+ model, a deep convolutional neural network (DCNN), to perform semantic segmentation on street view images. Subsequently, a random forest model was applied to integrate female safety perception scores with the street element dataset, constructing a gender-sensitive safety perception evaluation model to derive safety assessment results for women in street spaces. The DeepLabv3+ model enhances segmentation accuracy through multi-scale atrous convolutions and adopts an encoder–decoder architecture. It leverages DCNN-based feature extraction combined with an atrous spatial pyramid pooling (ASPP) module [58], enabling multi-scale contextual information capture and precise object boundary delineation. The model was trained on the ADE-20K dataset released by MIT, with segmentation performance evaluated using the mean intersection over union (mIoU) metric, a standard measure quantifying pixel-wise overlap between predictions and ground truth [59]. The trained model achieved an mIoU of 0.769. This process extracted the proportional coverage of environmental elements, such as sky, roads, vegetation, and buildings, within street scenes, forming the street element dataset (Figure 3).

The study then utilized a random forest model to couple female volunteers’ safety perception scores with the streetscape element dataset, establishing a female perspective safety evaluation model [34] to obtain safety assessment results for women in street spaces. The evaluation recruited 32 female volunteers to collect street safety perception data. To mitigate potential influences from factors such as socio-cultural background and spatial familiarity, volunteers were selected through interviews to ensure they had no prior experience of harassment and had never visited the study area. This process was conducted with the volunteers’ explicit consent, approved by the Ethics Review Committee, and no personal information was identified during the evaluation.

Given that human vision possesses a strong capacity for global attribute recognition, observers achieved 75% correct performance with presentations ranging from 19 to 67 ms, reaching maximum performance at about 100 ms. Therefore, the individual visual system enables relatively comprehensive scene perception and evaluation [60].

Prior to evaluation, volunteers were required to review all street view images. Evaluation is conducted from two perspectives: built environment safety and behavioral activity safety. The study used a Likert scale method, dividing safety perception evaluation into five levels: 0–19, 20–39, 40–59, 60–79, and 80–100 points, where higher scores indicated a stronger street safety. During the evaluation, photos were presented in a non-fully randomized order, with each photo displayed for 20 s. Volunteers were asked to evaluate the first 50 photos, and the evaluation framework generated predicted scores based on their ratings (Figure 4). Inspired by the iterative feedback module of cell phenotype scoring, this framework could autonomously learn and adjust predicted scores based on user behavior.

The model can fit the relationship between visual scene features and volunteer ratings: after volunteers complete safety evaluations, the model establishes a random forest ensemble to capture associations between environmental elements and individual scores. When volunteers rate subsequent images, the model learns patterns based on the relationship between their previous ratings and scene features, then recommends a score. Due to the black-box nature of the machine learning process, subsequent correlation analysis between street environments and sense of security is required to reveal underlying influencing mechanisms.

Prior to the formal evaluation, user pre-ratings were employed to validate the model’s accuracy. The study used 60% of the user pre-ratings for model training and 40% as a test set for validation. The fitting results indicated a mean error of 3.88%, RMSE of 5.06, OOB error of 11.37%, and OOB RMSE of 14.52, demonstrating a high level of consistency.

When the predicted scores for more than 5 photos deviated significantly (≥10 points) from the volunteers’ ratings, the random forest model would retrain and recalibrate the fitting model until a human–machine compromise state was achieved (where the difference between machine-predicted scores and volunteer ratings was within ±5 points), ultimately yielding the final female safety evaluation results.

Finally, the weights for the questionnaire survey and machine learning model evaluation results were determined using the expert scoring method. The final safety perception assessment result was obtained after calculating the combined outcomes of both approaches.

3.4. Construction of Stay Behavior Indicators

To objectively characterize female stay behaviors in street spaces, this study integrated field surveys and previous research to construct a stay behavior indicator system, categorizing stay behaviors into three dimensions: stay behavior density, stay duration, and stay behavior ratio (

Table 3. Stay Behavior Indicator System.

Primary Indicators	Secondary Indicators	Quantitative Explanation
Stay behavior	Stay behavior density (FSB1)	Staying population/street segment length Data.
	Stay duration	Short-term stay (FSB2)	Duration of stay is adopted as a key metric for staying activities, with 15 min intervals serving as the threshold: stays ≤15 min are classified as short-term stay, while stays >15 min are categorized as long-term stay.
		Long-term stay (FSB3)
	Stay behavior ratio(FSB4)	Staying population/total pedestrian population.

). Furthermore, women’s stay behaviors in streets can be subdivided into three types: commercial stay behavior, leisure stay behavior, and social stay behavior [16] (

Table 4. Taxonomy of Stay Behaviors.

Primary Indicators	Secondary Indicators	Main Content
Commercial stay behavior	Commercial hesitation	Viewing merchandise, street vendor merchandise
	Commercial stay	Inquiry, purchase, and queueing in retail environments
Leisure stay behavior	Leisure observation	Observing landmarks, buildings, and taking photography
	Leisure stay	Phone calls, waiting, sitting idle, organizing belongings
Social stay behavior	Social contact	Group conversational behaviors in public spaces, such as seated conversations, waiting-phase conversations
	Social entertainment	Group recreational activities in public spaces, such as parent–child play, group photography, live streaming

).

(1)

Stay behavior density translates stay behavior data to reflect the concentration of stay behaviors per unit street length over a specified period [61]. In this study, it refers to the density of women’s stay behaviors within a 100 m street segment over 30 min.

(2)

Stay duration measures the time length of individual stay behaviors. Based on prior studies, 15 min is adopted as the recording interval [26]. Stay durations <15 min are classified as short-term stays, while those ≥15 min are considered long-term stays.

(3)

Stay behavior ratio indicates the probability of stay behaviors occurring within a street segment. It is calculated as the ratio of women exhibiting stay behaviors to the total number of people in a 100 m street segment.

3.5. Structural Equation Modeling

Building upon the indicator framework, the model incorporates 5 latent variables and 22 observed variables to construct a structural equation model (SEM) examining the influence of street spatial female safety on women’s stay behaviors (Figure 5). In this model, women’s safety perception serves as the mediating variable, street environmental element indicators function as independent variables, and women’s stay behavior indicators act as dependent variables. Within the path diagram, dashed lines represent indirect effects, while solid lines represent direct effects. The specific computational formulas are as follows [46]:

$$\eta =B\eta +\Gamma \xi +\zeta$$

(1)

$$Y=\Lambda y+\epsilon$$

(2)

$$X=\Lambda x\xi +\delta$$

(3)

In the equation: observed variables X and Y are associated with their corresponding latent variables η and ξ through factor loadings Λy and Λx, respectively; ε and δ represent the residual terms of the measurement model; B denotes the coefficient matrix of effects between endogenous latent variables η; Γ represents the effects of exogenous latent variables on endogenous latent variables; and ζ stands for the residual term of the structural model.

4. Research Area and Data Source

4.1. Research Area

This study selected Dalian’s Xi’an Road area as the research site, located in the city core as a key transportation hub and commercial center. Preliminary surveys and situational analysis revealed that women’s stay behaviors predominantly occur in areas with dense commercial and infrastructural facilities. The Xi’an Road area features comprehensive infrastructure, high pedestrian flow, and concentrations of commercial and office facilities, encompassing various street types with a predominance of commercial streets.

Influenced by urban conditions in China, gated residential communities restrict access to non-resident vehicles and individuals. The street environments within these communities are homogeneous, and commercial functions are primarily concentrated on external streets. As a result, the perceived safety in such areas cannot represent urban street conditions broadly. Therefore, the study selected a research area centered on Xi’an Road with a 500 m walking-friendly radius, excluding gated residential zones (Figure 6). The blue points in Figure 6 represent data sampling locations spaced at 100 m intervals.

4.2. Data Source and Processing

This study sourced street data from the OpenStreetMap platform, supplemented and adjusted through field surveys. After topological processing of the road network, 146 street points were generated at 100 m intervals. Using Python tools (Python3.11.4), point of interest (POI) data within the study area was crawled. Subsequently, utilizing the open API of a map service, street view images were collected for each sampling point at four orientations (0°, 90°, 180°, and 270°), resulting in a total of 584 images. These images were processed through DeepLabv3+ for semantic segmentation, creating the Xi’an Road streetscape dataset.

(1)

Street environmental element indicators: Derived from the semantic segmentation streetscape dataset and POI data.

(2)

Female safety perception scores: The safety perception of street spaces in the Xi’an Road area was assessed using a questionnaire survey and a female safety perception evaluation model. For the questionnaire, 160 surveys were distributed, and after filtering, the valid response rate was 95.6%. Reliability and validity analysis showed strong results: Cronbach’s α = 0.853 (>0.7), KMO index = 0.782 (>0.7), and significance p = 0.000 (<0.05), confirming the survey’s statistical robustness. Using the expert scoring method, weights for built environment safety and behavioral activity safety were determined (

Table 5. Weighting Results.

Primary Indicators	Secondary Indicators	Measurement Method
Female Safety	Built Environment Safety (Weight: 0.624)	Questionnaire Survey (Weight: 0.492)
		Evaluation Model (Weight: 0.508)
	Behavioral Activity Safety (Weight: 0.376)	Questionnaire Survey (Weight: 0.612)
		Evaluation Model (Weight: 0.388)

). These weights were integrated with the results from the machine learning evaluation model to generate the final safety perception assessment.

(3)

Stay behavior indicators: After preliminary surveys, data collection was conducted on sunny weekday afternoons (14:00–15:00) through video recordings and behavioral annotations at 15 min intervals, forming a behavioral map of the area.

The study normalized the aforementioned indicator data using min–max scaling and employed Pearson correlation analysis in SPSS(SPSS 27) to examine the relationships between street environmental elements, women’s safety perception, and stay behaviors. Subsequently, a structural equation model (SEM) was constructed with female safety as the mediator, street environmental indicators as independent variables, and women’s stay behavior indicators as dependent variables to test mediation effects. During the model pre-construction phase, the factor loading coefficients of all variable indicators exceeded 0.7. Additionally, a multicollinearity test was conducted on the street indicators (

Table 6. Collinearity Diagnostics.

Indicators	Tolerance	VIF
Functional Density	0.431	2.32
Functional Mix	0.481	2.081
Pedestrianization Degree	0.458	2.182
Motorization	0.736	1.359
Visual Walkability	0.648	1.543
Interface Complexity	0.507	1.973
Sidewalk Area Ratio	0.649	1.541
Sidewalk Height Difference	0.624	1.602
Interface Transparency	0.522	1.917
Scenario Diversity	0.451	2.215
Green View Index	0.419	2.389
Sky View Factor	0.389	2.569
Enclosure	0.382	2.616
Spatial Congestion	0.604	1.656
Street Height-to-Width Ratio	0.493	2.03
Security Facilities Ratio	0.375	2.666
Traffic Signs Ratio	0.407	2.457

), and the results confirmed the absence of collinearity issues among the variables. The model was implemented in Amos using bootstrap resampling with 5000 iterations at 95% confidence intervals for robust mediation analysis. Among these, χ²/DF = 1.921, GFI = 0.803, TLI = 0.895, CRMR = 0.02, RMSEA = 0.065, and AGFI = 0.812, all of which meet the critical thresholds for model fit.

5. Results

5.1. Descriptive Statistics

5.1.1. Descriptive Statistics of the Sample

A total of 32 volunteers participated in the evaluation, with demographic statistics summarized in

Table 7. Descriptive statistics of the sample.

Variables	Descriptions	Number (Percentage)
Age	18–25 years	9 (28.1%)
	26–30 years	10 (31.3%)
	31–40 years	8 (25.0%)
	41 years and above	5 (15.6%)
Education Level	Junior high school or below	3 (9.4%)
	High school/vocational school	8 (25.0%)
	Bachelor’s/Associate degree	9 (28.1%)
	Master’s degree or above	12 (37.5%)
Income Range (CNY)	<5000	11 (34.3%)
	5000–7500	6 (18.8%)
	7500–10,000	10 (31.3%)
	>10,000	5 (15.6%)
Occupation	Student	11 (34.3%)
	Corporate employee	8 (25.0%)
	Civil servant/ public institution staff	10 (31.3%)
	Freelancer	3 (9.4%)

. The majority of volunteers (56.4%) were between 26 and 40 years old. Most participants were highly educated, with 65.6% holding a bachelor’s degree or higher. In terms of income, 50.2% earned between CNY 5000 and CNY 10,000 per month. Regarding occupations, enterprise employees, civil servants, and public institution staff constituted the largest group (56.4%), followed by students (34.4%).

5.1.2. Spatial Distribution of Street Environment Attributes and Female Safety Perception

The spatial visualization of street environmental indicators and women’s safety perception data (Figure 7 and Figure 8) reveals significant distributional heterogeneity across the Xi’an Road area. In terms of built environment safety, areas with high perceived safety are concentrated in the middle section of Xi’an Road, the intersection with Huanghe Road, and the junctions of Xinggong Street and Zhongchang Street (score range: 63.25–72.25). Regarding behavioral activity safety, high safety perception values are observed in the middle section of Xi’an Road and along Zhongchang Street.

Female safety demonstrates clustered spatial patterns rather than random distribution, with distinct high-value and low-value concentrations. High-value (63.000–71.00) clusters predominantly emerge in central Xi’an Road, Huanghe Road, Zhongchang Street, and Changxing Street, whereas low-value (48.00–57.25) zones concentrate in northern Lianhe Road, Minquan Street, and intersections of Guangping Street with Changli Street. Notably, areas with dense commercial facilities, complete infrastructure, and abundant public spaces correlate strongly with safety perception hotspots, while zones adjacent to aging residential areas with inadequate amenities exhibit lower safety scores.

Key environmental indicators, such as functional density (1.66–2.56), functional mix (1.849–2.073), pedestrianization degree (0.0697–0.1262), and security facilities ratio (0.0257–0.1552), demonstrate parallel spatial distributions with safety perception, all showing high-value concentration in central Xi’an Road. This confirms the area’s superior facility concentration and pedestrian-friendly design.

Conversely, sidewalk height differences, green view index, and traffic signs ratio distribution suggest generally flat terrain with sparse traffic signage across the study area. Urban greenery appears limited except at specific locations like Zhongchang–Changxing Street intersection, May Fourth Square, and Changxing–Wansui Street junction, where vegetation clusters correspond to green view index peaks.

5.1.3. Women’s Stay Behavior Data

Through on-site observation and recording at street points, data on women’s stay behaviors was obtained. From the perspective of behavior classification (Figure 9a), commercial and social stay behaviors dominate among women in the Xi’an Road area, primarily including window shopping, inquiring about or purchasing goods, queuing, and group conversations, accounting for 70–80% of all behaviors.

In terms of data distribution, commercial hesitation behaviors occupy a relatively large proportion, appearing across all street points (60–90%). Commercial stay behaviors are notably high in certain street sections (40–50%). Social contact behaviors are more scattered in distribution but reach higher proportions in specific locations (50–60%). Social entertainment behaviors exhibit extreme values in some street points (70–100%) and serve as the primary staying behavior for women.

The spatial distribution results of women’s stay behaviors (Figure 9b) show clear clustering characteristics. High-density concentrations appear in the middle section of Xi’an Road (3.70–9.57), while low-density clusters are observed near the intersection of Xinggong South Fifth Street and Zhongchang Street, Huanghe Road, Shengping Street, and Yong’an Street (0.20–1.30). The high-density areas may be attributed to the concentration of commercial and facility points, along with well-developed supporting infrastructure and high pedestrian friendliness. The spatial distributions of stay behavior ratio, short-term stays, and long-term stays are similar, all exhibiting a distinct core–periphery structure within the region, with high-value areas concentrated in the middle section of Xi’an Road. However, short-term and long-term stays have different high-value core points: near Ruyi Street (0.4761–0.5230) and near Wansui Street (0.1219–0.1428), respectively.

5.2. Correlation Analysis

The correlation between street environmental factors and women’s safety and stay behaviors is shown in Figure 10 and Appendix A, along with an analysis of the relationships among different variables. The findings on how street environments influence women’s sense of safety reveal that 13 environmental variables have significant effects. Among them, functional density, pedestrianization degree, visual walkability, sidewalk area ratio, sky view factor, safety facilities ratio, and traffic signs ratio exhibit positive correlations with female safety. Conversely, functional mix, motorization, scenario diversity, enclosure, spatial congestion, and street height-to-width ratio show negative correlations with women’s sense of safety.

The correlation analysis between street environmental factors and women’s stay behaviors indicates the following relationships: in terms of women’s stay behavior ratio, functional density, pedestrianization degree, visual walkability, green view index, sky view factor, security facilities ratio, and traffic facilities ratio show positive correlations, while sidewalk height differences, scenario diversity, enclosure, and street height-to-width ratio exhibit negative correlations. Regarding women’s stay behavior density, functional density, pedestrianization degree, visual walkability, sidewalk area ratio, green view index, sky view factor, and safety facilities ratio demonstrate positive correlations, whereas scenario diversity and enclosure display negative correlations. For short-term stays among women, functional density, pedestrianization degree, visual walkability, green view index, sky view factor, safety facilities ratio, and traffic facilities ratio are positively correlated, while functional mix, motorization, interface complexity, sidewalk height differences, scenario diversity, enclosure, spatial congestion, and street height-to-width ratio are negatively correlated. Concerning women’s long-term stays, functional density, pedestrianization degree, visual walkability, sidewalk area ratio, green view index, sky view factor, and safety facilities ratio present positive correlations, and scenario diversity and enclosure show negative correlations.

The concentration level of surrounding facilities, pedestrian-friendly street design, proportion of natural elements, and ratio of safety facilities collectively contribute to a safer pedestrian environment for women, while increasing both the occurrence rate and duration of women’s stay behaviors. Conversely, factors such as motor vehicle traffic volume, complexity of street interface elements, and spatial enclosure may diminish perceived safety and reduce the likelihood of stay behaviors.

5.3. Mediation Effect Analysis

The constructed structural equation model was calibrated and adjusted for errors to obtain operational results. Regarding model fit indices (

Table 8. Model fitting information.

	χ2/DF	GFI	TLI	CRMR	RMSEA	AGFI
Model	1.921	0.803	0.895	0.02	0.065	0.812
Reference [55]	<3	>0.7	>0.7	<0.08	≤0.08	>0.7

), the χ²/DF, GFI, TLI, and RMSEA all met their respective critical thresholds, indicating that the structural equation model demonstrates satisfactory computational performance and good fit. Furthermore, both point estimates and interval estimates of the mediation effects were obtained. The 95% confidence intervals excluded zero, demonstrating statistically significant mediation effects and providing support for the study’s fundamental hypotheses.

According to the model results (Figure 11,

Table 9. Model Results.

Influence Pathway	Estimate	Standardized Estimate	S.E.	C.R.	p
Functionality → female safety (H1a)	0.314	0.371	0.062	5.085	***
Interface morphology → female safety (H1b)	0.168	0.212	0.056	3.011	0.003 **
Spatial quality → female safety (H1c)	0.238	0.304	0.059	4.061	***
Street facilities → female safety (H1d)	0.253	0.317	0.059	4.265	***
Female safety → stay behavior (H2)	0.417	0.368	0.118	3.541	***
Functionality → stay behavior (H3a)	0.282	0.294	0.082	4.199	0.002 **
Interface morphology → stay behavior (H3b)	0.220	0.236	0.072	3.382	***
Spatial quality → stay behavior (H3c)	0.152	0.176	0.069	2.074	0.01 *
Street facilities → stay behavior (H3d)	−0.082	−0.109	0.071	−1.216	0.224

Note: * Relationship is significant at the 0.05 level, ** Relationship is significant at the 0.01 level, *** Relationship is significant at the 0.001 level.

), among street functionality characteristics, visual walkability (F5) and functional density (F1) demonstrated the highest contribution degrees at 0.842 and 0.819, respectively, followed by progressively decreasing influences from motorization level (F4), functional mix (F2), and pedestrianization degree (F3). Functional density reflects the agglomeration level of various facilities surrounding the street, while visual walkability quantitatively measures street pedestrian-friendliness from a walker’s perspective. These two indicators substantially shape the pedestrian usage environment of street spaces.

Among street interface morphology indicators, scenario diversity (IM5) and interface transparency (IM4) exhibit the strongest influence with coefficients of 0.852 and 0.814, respectively, followed by progressively weaker contributions from sidewalk area ratio (IM2), interface complexity (IM1), and sidewalk height difference (IM3). These two dominant factors, scenario diversity and interface transparency, quantitatively represent the richness of street facilities and the visual penetration capacity for pedestrians, effectively characterizing the complexity and diversity of street spatial interface morphology.

The modeling results demonstrate that, among street spatial quality indicators, the street height-to-width ratio (SQ5) exerts the most substantial positive influence (0.856), followed by progressively decreasing effects from sky view factor (SQ2), spatial congestion (SQ4), enclosure (SQ3), and green view index (SQ1). Notably, both street height-to-width ratio and sky view factor show significant correlations with pedestrians’ spatial quality perception, ultimately influencing their behavioral patterns, psychological responses, and activity participation levels within street environments.

Among street facility factors, safety facilities ratio (SF1) demonstrates the strongest direct positive effect (0.946) and plays a dominant role, followed by traffic signs ratio (SF2). Regarding women’s stay behaviors, stay behavior ratio (FSB4) and stay behavior density (FSB1) emerge as the most influential factors, showing significant positive contributions with path coefficients of 0.937 and 0.794, respectively, while both short-term stays and long-term stays exhibit relatively weaker effects.

According to the significance test results of the mediating effect (Figure 11), female safety plays varying degrees of mediating and facilitating roles in the influence pathways of street functionality, interface morphology, and spatial quality on women’s stay behavior. Moreover, it serves as a complete mediator in the impact of street facilities on women’s stay behavior. Specifically, when the street spatial environment influences female staying, sense of security plays a significant role in this pathway and becomes a core psychological demand for women when choosing to stay. If the street environment triggers psychological unease, women may reduce their stay behavior to minimize risk exposure, making sense of security a subjective bridge connecting urban environments and behavior. In the influence path of street facilities on women’s stay behavior, the complete mediating role of safety may stem from the fact that street facilities themselves do not directly determine stay behavior but indirectly affect women’s staying decisions through this mediating mechanism. Only when street facilities are well-equipped and provide sufficient security are women more likely to engage in staying.

In the path of latent variables’ influence on the mediating variable, street functionality and street facilities exhibit the strongest positive effects on women’s safety, with coefficients of 0.371 and 0.317, respectively. Spatial quality has a moderate influence (0.304), while interface morphology shows a relatively weaker positive effect (0.212). Specifically, the street functionality indicator describes street-level diversity, commercial clustering density, and pedestrian-friendly design, whereas the street facilities indicator focuses on the completeness of infrastructure and the rationality of spatial distribution. The results suggest that, when a street combines diverse functional formats with well-equipped facilities, female safety is significantly enhanced. Additionally, the findings indicate potential synergistic effects among street functionality, street facilities, and spatial quality, while interface morphology plays a supplementary role, albeit with a comparatively limited impact.

In the path of the mediating variable’s influence on women’s stay behavior, sense of security contributes a significant positive effect and serves as the core mediating factor driving women’s staying decisions. Due to the dual influence of biological differences and social roles, women are more sensitive to safety demands in public spaces than men, meaning their safety perception directly shapes spatial usage decisions.

In the pathway of latent variables’ influence on women’s staying behavior, street functionality, interface morphology, and spatial quality all demonstrate significant positive effects, with functional formats showing the strongest contribution (0.294), while street facilities exhibit no significant impact. Specifically, functionality reflecting street-level diversity, commercial clustering density, and pedestrian-friendliness directly fulfills women’s daily travel and lifestyle needs, establishing a strong demand-driven behavioral linkage. In contrast, interface morphology and spatial quality relate more to women’s spatial comfort experience, forming a secondary influence mechanism. Although street facilities can enhance security, women may perceive them as basic prerequisites rather than direct motivators for staying, resulting in their negligible effect on behavioral decision making.

6. Discussion

6.1. Discussion of Influence Mechanisms

6.1.1. The Impact of Urban Street Environment Attributes on Women’s Safety Perception and Stationary Behavior

The study validates the influence pathways of environmental factors on individual perception and behavior as posited by the lens model theory. Streets, as complex systems composed of multiple environmental elements, exert asymmetric effects on female safety and stay behavior. During cognitive processing and behavioral decision making, individuals selectively perceive different environmental cues, with certain elements generating either positive or negative impacts. From a holistic cognitive perspective, spatial perception represents a subjective synthesis formed through the cognitive system processing of multiple urban environmental elements and their structural relationships.

Functional density reflects the richness and clustering intensity of various facilities in street environments. While high functional mix may represent complexity and uncertainty in certain contexts, moderate functional density indicates the availability of diverse life services and activity spaces. This helps attract and sustain pedestrian flow, creating continuous natural surveillance [62] which places women within a “visible” safety network and significantly enhances their sense of security. Furthermore, appropriate functional density can transform streets from transitional spaces into destination spaces, reinforcing the legitimacy of behavior and the sense of territoriality [61] while increasing the likelihood of stay activities.

Environmental indicators related to walking behavior measure the extent to which a street supports pedestrian activity. Streets with high walkability typically feature appropriate walking scales, continuous pedestrian pathways, and well-maintained walking facilities. Continuous and legible walking spaces can reduce cognitive load for women, making it easier to identify routes, locate directions, and predict environmental conditions. This significantly enhances their sense of security and increases their willingness to stay—a finding supported by existing research [62].

Natural element indicators in street environments, such as green view ratio and sky openness, reflect environmental quality and comfort levels. According to attention restoration theory [63], abundant greenery and open skies provide visual “rest points” that effectively alleviate mental fatigue and psychological stress caused by overstimulating urban environments—such as dense crowds and noisy traffic—thereby fostering positive safety feedback and a greater willingness to stay.

Street facilities reflect the quantity and distribution of safety infrastructure and traffic signage. Well-equipped street facilities serve as a critical foundation for street safety, which is consistent with the findings of this study. Safety amenities such as surveillance cameras and streetlights provide real-time security assurance for women. Adequate lighting and monitoring systems not only meet basic functional needs but also establish prerequisites for natural surveillance. By eliminating visual obstructions and shadows, they significantly enhance environmental visibility and permeability. Traffic signs and signals (e.g., traffic lights and guide signs) offer clear behavioral guidance and wayfinding options, reducing environmental uncertainty and navigation stress, thereby strengthening women’s spatial sense of security. This aligns with the conclusions of Xu Leiqing et al. [64] in their study on safety perception in public spaces.

6.1.2. The Role of Safety Perception in the Pathway from Street Environment to Women’s Stationary Behavior

The research validates the mediating and moderating role of perception in the relationship between environment and individual behavior as hypothesized in spatial cognition theory. Women’s daily activities occur within physical spaces composed of various urban environmental elements, where environmental influences on individual behavior are typically mediated through cognitive mechanisms. As a cognitive and symbolic processing of environmental information, perception serves as a crucial factor through which objective urban environments affect individual behavior. From the perspective of individual spatiotemporal behavior, urban space emerges within specific planning and policy contexts, grounded in spatial perception. It constitutes a multi-dimensional spatial entity with distinct morphological configurations and structural compositions, formed through the interplay of natural or built environments, social activities, and subjective perceptions across different scales and scenarios [65].

In urban street environments, female stay behavior is influenced by multiple factors, with sense of security serving as the key mediating variable connecting environmental factors with behavioral decisions. Specifically, street environmental elements affect women’s staying decisions by influencing their safety perception. When particular environmental elements trigger psychological discomfort, women tend to actively reduce potential risk exposure by decreasing staying frequency or shortening duration, following a chain-reaction mechanism of “street environment recognition–safety perception–behavioral adjustment in staying.” Changes in environmental elements may thus alter safety perception and consequently modify women’s staying decisions. Among these factors, street functionality, reflecting street service capacity, facility distribution, and pedestrian-friendliness, carries greater weight in the “street environment–female safety–stay behavior” pathway, while spatial quality and interface morphology show relatively secondary effects. Notably, street facilities themselves do not directly influence stay behavior, they must first be evaluated through women’s safety perception before being translated into actual staying decisions.

6.2. Optimization Strategies

Based on the above discussion regarding the influence of street environments on women’s safety, stay behaviors, and the mediating role of perceived safety, the following street space planning strategies are proposed to foster high-quality urban street spaces from a female-friendly perspective (

Table 10. Strategies.

Strategies	Detailed Implementation Strategies
Optimize the functional layout of streets to create refined anchor points and vibrant spaces	New urban districts	Implement the “small block, dense street network” model to encourage street-level placement of retail, cafés, convenience services, and other amenities frequently used by women.
	Older urban districts	Utilize scattered green spaces or idle corners at street intersections to create “pocket parks” or mini-plazas. Introduce movable urban furniture such as: Planting boxes that integrate greening and spatial definition; multi-level terraces offering seating, resting areas, and support for small-scale events; movable tables, chairs, and sunshades to accommodate temporary social or work needs.
Improve the pedestrian system to create recognizable and female-friendly comfortable routes	New urban districts	Maintain a street height-to-width ratio between 1:1 and 1:1.5.
	Older urban districts	Replace solid walls at the ground-level building interface with transparent display windows and full-height glass doors. Extend interior functions of ground-floor spaces (e.g., cafés, bookstores, retail shops) outward by adding outdoor seating areas. Introduce corner buffer zones at mid-block segments to enhance spatial transition and safety. Use ground markings, colored pavements, and subtle topographic variations to guide pedestrian flow and define walking paths.
Enhance street environmental quality and establish visible support facilities for women	New urban districts	The green view index (GVI) along primary residential streets should be no less than 25%. Building setbacks must be regulated to ensure a sky view factor (SVF) of at least 30%.
	Older urban districts	Introduce planting boxes and supplement street tree planting. Lighting design should ensure an average horizontal illuminance of no less than 15 lux on pedestrian walkways, with a uniformity ratio above 0.4, to avoid significant shadow areas. Surveillance cameras should cover key public spaces and path intersections, installed at a height of 2.5–3 m, and feature visible signage to indicate their presence.
Establish a gender-inclusive street design and governance system	During the preliminary research, design development, and post-evaluation stages of the project, focus groups comprising women of diverse ages and professional backgrounds should be organized to conduct on-site walk-through audits (Walk Audits).

).

(1)

Optimize the functional layout of streets to create refined anchor points and vibrant spaces.

Transform “transitional” streets into attractive and belonging-rich “destination” spaces. Research findings indicate that moderate functional density significantly enhances women’s sense of security and frequency of stationary behaviors.

In the planning of new urban districts, planning authorities should implement the “small block, dense road network” model and restrict large single-functional parcels. Clearly define mixed-use ratios for commercial, service, and cultural facilities, and encourage street-level of retail, cafés, convenience services, and other high-frequency-use amenities for women. For the renovation of streets in older urban areas, utilize scattered green spaces or idle corners at street intersections to create “pocket parks” or mini-plazas. These spatial anchors can attract spontaneous gathering and lingering among women, thereby activating the entire street segment. Design interventions may introduce movable urban furniture, including: planters that serve both greening and spatial definition functions; multi-level terraces that provide seating and resting areas and support small-scale events; movable tables, chairs, and sunshades to accommodate temporary social or work needs (Figure 12 and Figure 13).

In studies on Barcelona’s neighborhood transformations [66], the introduction of modular, low-cost, and easily adaptable street furniture in underutilized corner spaces has effectively revitalized street environments and significantly increased pedestrian retention rates.

(2)

Improve the pedestrian system to create recognizable and female-friendly comfortable routes.

Ensure physical and visual continuity of streets by eliminating discontinuities and breakpoints. Research results demonstrate that continuous and legible pedestrian spaces reduce cognitive load for women, thereby positively influencing their safety and stay behavior. Therefore, urban planners should regulate street height-to-width ratios to create pleasant and continuous street spatial scales, enhance ground-floor building interfaces, and activate street-level commercial functions. Additionally, pedestrian cross-sections should be utilized to incorporate rest and activity areas.

In the planning of new urban districts, planning authorities should control the street height-to-width ratio within the range of 1:1 to 1:1.5 to create a comfortable street atmosphere. For the renovation of streets in older urban areas, facade improvements can be implemented to enhance interface permeability. Property owners should be encouraged to replace solid walls at the ground level with transparent display windows and full-height glass doors to improve visual connectivity. Additionally, interior functions of ground-floor spaces (such as cafés, bookstores, and retail shops) can be extended outward by adding outdoor seating areas. This kind of “indoor–outdoor integration” design injects the vitality of private interiors into the public realm, providing abundant “eyes on the street” and continuously activating the streetscape (Figure 14 and Figure 15). Observational studies conducted in commercial streets in Shanghai and Da Nang [67] have shown that transparent glazing helps extend visitors’ dwell time, enhances visual engagement with the street, and effectively increases women’s sense of security.

Furthermore, in Barcelona’s block renovation initiatives [68], corner buffer zones have been introduced at mid-block segments between intersections to eliminate street discontinuities. Pedestrian pathways are guided through measures such as ground markings, colored pavements, and subtle topographic variations, thereby improving the walking system throughout the street network.

(3)

Enhance street environmental quality and establish visible support facilities for women.

Planners can utilize natural elements to alleviate urban stress and provide visual and psychological rest points for women. Research findings indicate that spatial quality and street facilities positively influence women’s sense of security and behavioral patterns. Therefore, planning strategies should regulate building density and height on both sides of streets to ensure adequate sky openness. Additionally, public safety authorities should ensure even distribution of safety facilities such as lighting and surveillance cameras, minimize visual blind spots for women, and provide clearly designed, legible signage systems.

In the planning of new urban districts, primary residential streets should maintain a green view index (GVI) of no less than 25% [69,70]. Climate-appropriate street tree belts should be incorporated to enhance shading and ecological function. Building setbacks must be regulated to ensure a sky view factor (SVF) of at least 30%, [67] avoiding the oppressive “canyon effect” in street corridors. For the renovation of older urban areas, continuous tree canopies should be established through the addition of planting boxes and supplemental street trees (Figure 16).

In the practice of Vienna’s Frauen-Werk-Stadt (Women-Work-City) project [66], the number of volleyball and badminton courts in neighborhood parks was significantly increased to counter the dominance often asserted by boys in basketball-dominated spaces. This approach effectively promoted sports participation among girls. Dedicated zones for resting, socializing, and observation were introduced around sports facilities, and safety was enhanced through improved lighting and widened sidewalks.

While street lighting and surveillance facilities can enhance women’s sense of security in urban areas, a study conducted in Xiamen’s commercial district [53] revealed that excessive monitoring equipment and signage systems may instead reduce pedestrians’ perceived safety, potentially evoking fear of unknown events. Therefore, the deployment of street facilities such as cameras and signboards should be moderate in quantity, focused on covering key public spaces and path intersections, and installed at an appropriate height of 2.5–3 m.

(4)

Establish a gender-inclusive street design and governance system

Female safety audit tools can be introduced to comprehensively capture women’s spatial experiences in streets. During the preliminary research, design development, and post-evaluation stages of a project, focus groups comprising women of diverse ages and professional backgrounds should be organized to conduct walking audits (Walk Audits). Through their firsthand perceptions, potential safety hazards and design flaws can be identified. This participatory design approach compensates for the limitations of purely quantitative data (e.g., machine learning ratings) by capturing hard-to-quantify emotional responses and experiential nuances, ensuring that design decisions genuinely address users’ needs. Furthermore, it promotes collaboration among urban planning, transportation, green space management, and public security departments to integrate gender-sensitive principles into the entire process, from macro-level planning to micro-level management.

7. Conclusions

7.1. Conclusions

This study establishes an influence pathway of “street environment–female safety–stay behavior” based on the spatial cognition hypothesis and lens model theory. Through correlation analysis and structural equation modeling, we examine how street environmental elements affect women’s sense of security and stay behavior, as well as the mediating role of security perception in this pathway. The findings reveal that: (1) different street elements exert varying degrees of positive or negative influence on female safety and stay behavior; (2) sense of security serves as a significant mediator in the environment–behavior pathway, directly affecting women’s staying decisions. Among street environmental factors, functionality demonstrates the strongest influence in this pathway, followed by spatial quality and interface morphology in descending order of impact, while street facilities exert only indirect effects through security perception.

7.2. Research Deficiency

This study has several limitations: (1) spatial perception needs can be categorized into a basic level (safety and comfort), intermediate level (interest and mystery), and advanced level (belonging and identity) [25]. While the study focused on fundamental perception needs that significantly influence staying, it did not examine the mediating effects of other perceptual dimensions. (2) The study’s data collection on women’s stay behaviors did not cover all time periods (e.g., day/night, weekends), which may limit the generalizability of the findings across different temporal and spatial contexts due to potential variations in environmental and behavioral patterns. (3) In assessing women’s sense of security, this study was limited by the quantitative nature of the machine learning approach and did not incorporate longitudinal tracking methods to obtain evaluation results. Future research could employ safety audit tools and text analysis models to conduct fine-grained analysis of women’s safety perception processes, enabling deeper exploration of long-term temporal patterns in perceived safety within street environments and individual coping mechanisms. (4) When analyzing environmental element impacts on perception and behavior, the study only considered their correlations and positive/negative directional effects. Existing research suggests threshold effects may exist for environmental factors’ influence [12,20,54,71,72], indicating future studies could employ more precise models to accurately determine each element’s impact parameters.