Consumer Perceptions and Attitudes Towards Ultra-Processed Foods

Abstract

The consumption of ultra-processed foods (UPFs) has become a central topic in discussions surrounding public health, nutrition, and consumer behaviour. This study aimed to investigate the key factors shaping customer perceptions and attitudes towards UPFs and explore their impact on purchase decisions. A total of 290 completed questionnaires from an online survey were analysed to identify the drivers influencing consumer actions and habits. Users’ opinions were systematised based on their attitudes towards UPFs, considering factors such as health consciousness, knowledge, subjective norms, and environmental concerns. Participants were then categorised using both traditional and advanced data analysis methods. Structural equation modelling (SEM), machine learning (ML), and multi-criteria decision-making (MCDM) techniques were applied to identify hidden dependencies between variables from the perspective of UPF consumers. The developed models reveal the underlying relationships that influence acceptance or rejection mechanisms for UPFs. The results provide specific recommendations for stakeholders across the food production and marketing value chain. Public health authorities can use these insights the findings to design targeted interventions that promote healthier food choices. Manufacturers and marketers can leverage the findings to optimise product offerings and communication strategies with a focus on less harmful options, aligning more closely with consumer expectations and health considerations. Consumers benefit from enhanced product transparency and tailored information that reflects their preferences and concerns, fostering informed and balanced decision-making. As attitudes toward UPFs evolve alongside changing nutrition and consumption patterns, stakeholders should regularly assess consumer feedback to mitigate the impact of these harmful foods on public health.

1. Introduction

Ultra-processed foods (UPFs), also referred to as industrially formulated food products, ready-to-eat convenience foods, or highly processed food items, have gained prominence in modern nutrition due to their affordability, extended shelf life, and convenience [1]. Initially, food processing was primarily aimed at preserving food and enhancing its safety. However, with the rise of industrial food production, an increasing number of food products contain additives, emulsifiers, artificial flavours, and preservatives, raising concerns about their impact on human health and well-being. Nowadays, UPFs account for a significant portion of global food consumption, displacing healthier unprocessed foods and influencing consumer preferences [2].

The growing prevalence of UPFs has sparked debates among policymakers, nutritionists, and consumers regarding their long-term effects on health. Studies suggest that the frequent consumption of UPFs is linked to various non-communicable diseases, including obesity, cardiovascular issues, type 2 diabetes, and metabolic disorders [3]. In addition to health risks, the environmental impact of UPFs, such as increased carbon footprints due to industrial food processing and packaging, has also drawn attention [4]. As a result, understanding consumer attitudes toward UPFs has become essential for shaping dietary guidelines, public health policies, and sustainable food systems.

According to recent market analyses, the demand for UPFs continues to grow [5], driven by factors such as rapid urbanisation, changing lifestyles, and aggressive food marketing strategies. The global UPF market has witnessed a steady increase in sales, particularly in middle- and high-income countries where convenience and affordability drive purchasing decisions [6]. Food companies continue to expand their product lines by introducing reformulated UPFs with reduced sugar, salt, and fat content, catering to health-conscious consumers while maintaining their market dominance [7].

Despite the public significance of the problem, there is no unified framework for studying consumer attitudes towards UPFs and their impact on dietary choices. Understanding the factors that shape consumer perceptions of UPFs and predicting their influence on purchasing behaviour is a complex issue due to three main reasons:

• The growing public debate surrounding UPFs involves health concerns, socio-economic factors, and environmental sustainability, all of which contribute to shifting consumer preferences [8]. For example, increased awareness of nutrition-related diseases and sustainable eating habits may lead consumers to reconsider their food choices, while economic constraints and lifestyle convenience continue to increase demand for UPFs.
• Although aggressive marketing, including digital advertising and celebrity endorsements, influences consumer trust and willingness to purchase these products [9], consumers are more mindful of the harmful consequences of consuming such foods. As a result, manufacturers have leveraged advancements in food technology and marketing strategies to offer and promote UPFs with healthier components and reduced nutrients of concern.
• The vast amount of data and the multifaceted nature of the problem require the application of innovative data analysis methods. Meanwhile, methods for analysing consumer attitudes and decision-making processes have expanded significantly with the integration of big data analytics, multi-criteria decision-making (MCDM), sentiment analysis, and other advanced artificial intelligence (AI) techniques [10]. These methods, tools, and technologies offer deeper insights into behavioural trends and help uncover the underlying motivations behind UPF consumption.

These challenges motivate us to examine consumer attitudes towards UPFs using both traditional statistical techniques and modern analytical methods, such as structural equation modelling (SEM) and MCDM models.

The objective of this study was to develop and validate comprehensive models that evaluate and predict consumer attitudes toward UPFs by considering key behavioural factors such as perceived health conciseness, food knowledge, subjective norms, and environmental awareness. The models also incorporate demographic and socio-economic factors, including income levels, gender, education, dietary habits, and consumer preferences in food shopping. This structured approach enhances the understanding of the determinants of UPF consumption and their implications for public health policies and food industry strategies.

The main tasks of this study were as follows:

• Develop a conceptual model that allows for the systematic examination of consumer attitudes towards UPFs and the identification of underlying behavioural patterns influencing purchasing decisions.
• Gather and structure a dataset reflecting consumer experiences, perceptions, and preferences regarding UPFs, incorporating key socio-economic and demographic factors, and food preferences.
• Identify the main determinants that impact consumer willingness to purchase UPFs by reviewing previous studies and proposing appropriate analytical approaches to assess their impact.
• Construct and validate mathematical models based on the identified factors and compare the results with findings from prior research on food consumption behaviour.

This study employs SEM to examine data from 290 survey participants, exploring the impact of health awareness, taste preferences, and social factors on consumer attitudes toward UPFs. By combining statistical analysis with modern decision-making approaches, the research offers valuable insights for policymakers, the food industry, and sustainability experts aiming to promote more balanced and healthier dietary choices.

The structure of this paper is as follows: Section 2 examines the main characteristics of UPFs and their significance in contemporary dietary patterns. Section 3 provides an overview of existing research on consumer perceptions of UPFs, emphasising the key psychological, socio-economic, and behavioural determinants of food choices. Section 4 outlines the questionnaire design, specifying the measurement indicators adapted from prior studies on food consumption and health awareness. Section 5 focuses on data analysis; SEM and AI methods are applied to develop and validate mathematical models, followed by a comparative assessment with findings from previous research. Lastly, Section 6 summarises the main conclusions, acknowledges the study’s limitations, and suggests potential directions for future investigations.

2. State-of-the-Art Review of Consumer Attitudes Towards Ultra-Processed Foods

UPFs are now deeply embedded in modern dietary patterns, influenced by their widespread availability, cost-effectiveness, and extended shelf life. While UPFs offer a convenient food option for many consumers, increasing concerns about their potential health risks have fuelled academic and public debates. Several studies indicate that the frequent consumption of UPFs is linked to negative health outcomes, including cardiovascular diseases, and metabolic disorders [11,12]. Additionally, the environmental impact of large-scale UPF production raises further concerns about sustainability and food system resilience [13].

Despite increasing awareness of these issues, consumer demand for UPFs remains strong, driven by aggressive marketing strategies, evolving food technologies, and shifts in lifestyle preferences. The rise of health-conscious consumers has prompted food manufacturers to reformulate UPFs by reducing unhealthy additives or incorporating functional ingredients, such as added probiotics, plant-based proteins, or fortified micronutrients [14]. However, public scepticism about the actual health benefits of these reformulated products persists. Consumers’ attitudes towards UPFs are dependent on various factors, including health awareness, price sensitivity, trust in food labelling, and socio-economic conditions.

2.1. Key Features and Taxonomy of Ultra-Processed Foods

The increasing presence of UPFs in modern diets has been propelled by advancements in food processing technology, evolving consumer lifestyles, and intensive marketing campaigns. The convenience, affordability, and extended durability of UPFs make them a preferred choice for many consumers despite growing concerns regarding their impact on health and well-being [15].

Recent trends in UPF consumption reflect several key developments:

• The expansion of functional UPFs—in response to increasing health consciousness, food manufacturers are reformulating UPFs by incorporating added vitamins, fibre, probiotics, and protein-enriched alternatives [9]. These so-called “healthier” UPFs aim to appeal to consumers looking for convenient yet nutritionally enhanced options.
• Growth of plant-based UPFs—the demand for plant-based diets has led to a surge in UPFs marketed as vegetarian or vegan alternatives, such as meat substitutes and dairy-free products. While these products align with sustainability and ethical consumption trends, they often remain highly processed, containing emulsifiers and synthetic ingredients [16].
• Ultra-convenience in food innovation—the rise of ready-to-eat meals, instant snacks, and meal replacement products reflects a shift in consumer preferences toward faster and more effortless eating solutions. Many of these products prioritise convenience over nutritional value, contributing to an increased intake of ultra-processed foods [17].
• Sustainability challenges and reformulation efforts—concerns over environmental sustainability have prompted some manufacturers to explore eco-friendly packaging, reduce food waste, and develop “clean-label” UPFs with fewer artificial additives [18]. However, balancing sustainability with affordability and profitability remains a challenge.
• Increased reliance on digital food marketing—brands leverage social media platforms, food delivery apps, and personalised advertising to target consumers with UPF promotions [19]. The use of influencer endorsements and algorithm-driven recommendations have contributed to the growing acceptance and appeal of UPFs, particularly among younger consumers.

Ultra-processed foods exhibit distinct characteristics that influence consumer choices and dietary patterns. They can be categorised based on multiple criteria, each reflecting different aspects of their composition, processing methods, and market positioning.

Nutritional profile: UPFs typically contain high levels of undesirable components while lacking essential nutrients such as fibre, protein, and vitamins. Some categories of UPFs have been reformulated to include functional ingredients [20].

Degree of processing: UPFs can be classified based on the intensity of industrial processing involved in their production. Some undergo extensive chemical modifications, such as hydrogenation or extrusion, while others rely primarily on artificial preservatives and stabilisers to enhance shelf life and taste. This classification helps differentiate between minimally processed, processed, and ultra-processed food categories [21].

Marketing and branding strategies: The promotion of UPFs heavily depends on digital marketing, social media advertising, and influencer endorsements. Companies use targeted advertising campaigns, leveraging consumer behaviour data to increase brand visibility and sales [22]. Additionally, the rise of plant-based and “healthier” UPFs reflects shifting consumer preferences influenced by wellness trends and environmental concerns.

Consumption context: UPFs can also be categorised based on their role in daily nutrition. Some are designed as meal replacements, such as protein bars and instant soups, while others function as snacks or indulgent treats, like packaged sweets and processed meat products. The convenience and affordability of these products make them appealing across different demographic groups [23].

Sustainability considerations: Another classification criterion relates to the environmental impact of UPF production. Some manufacturers are adopting eco-friendly packaging, reducing food waste, and reformulating products to align with sustainability goals. However, the industrial-scale production of many UPFs continues to contribute to resource depletion, greenhouse gas emissions, and plastic pollution [13].

The categorisation of ultra-processed foods provides valuable insights for nutritionists, public health officials, and policymakers in shaping dietary guidelines and regulatory frameworks. Understanding the different types of UPFs and their appeal to consumers is essential for designing effective interventions that encourage healthier food choices while considering economic and lifestyle constraints.

2.2. Assessing Ultra-Processed Foods

Evaluating and comparing UPFs requires a structured approach that considers multiple dimensions of their impact on health, consumer behaviour, and the environment. Various assessment tools can be applied to analyse UPFs, broadly categorised into three main areas: nutritional metrics, composite indices, and theoretical models. These instruments provide valuable insights for industry stakeholders and policymakers in better understanding and regulating UPF consumption.

Nutritional metrics: UPFs are often evaluated based on their composition, including levels of potentially harmful components, and artificial additives. Standardised scoring systems, such as the NOVA classification [21] help rank foods based on their degree of processing and nutritional quality. These tools enable consumers and health professionals to identify and compare the potential health risks associated with different UPFs. These metrics for food classification can also be applied to evaluate the health implications of UPFs, helping consumers and policymakers make informed dietary choices.

Composite indices: Several compound indices integrate multiple factors to assess the overall health impact of UPFs such as Nutri-Score [24], Ofcom [25], and Health Star Rating (HSR) [26]. These indices often combine nutritional data with other key elements, such as glycaemic load, caloric density, and the presence of synthetic additives. In addition, consumer perception indices measure factors like taste, affordability, and brand trust, influencing UPF purchasing behaviour [27,28]. Despite the growing use of these indices, there is currently no universal standard for assessing UPFs. While organisations like the World Health Organization (WHO) and the Food and Agriculture Organization (FAO) provide general dietary guidelines, there is no globally recognised framework for systematically evaluating the health impact of ultra-processed foods. Moreover, emerging research continues to refine these indices, aiming to provide more accurate and actionable insights into food quality and its implications for public health.

Theoretical models: Conceptual models provide a broader perspective on UPF consumption by incorporating psychological, social, and economic factors. Frameworks such as the Health Belief Model (HBM) [29] and the Theory of Planned Behaviour (TPB) [30] help explain consumer attitudes and decision-making regarding UPFs. Additionally, MCDM techniques and ML models have been applied to analyse the complex interactions between health awareness, convenience, and purchasing preferences. These theoretical models provide structured approaches to understanding UPFs’ nutritional and behavioural impact, offering valuable insights for developing public health policies and consumer education strategies.

By integrating these different assessment tools, researchers and policymakers can develop a more comprehensive understanding of UPFs and their role in modern diets.

3. Related Work

3.1. Consumer Attitudes Towards Ultra-Processed Foods and Their Influence on Purchase Intentions

Understanding consumer attitudes towards UPFs is essential for both businesses and policymakers seeking to balance market demands with public health objectives. By analysing the psychological, social, and economic factors that drive UPF consumption, this study aims to provide insights into how consumer decision-making processes align with broader health and sustainability trends. The findings will help guide targeted marketing approaches, regulatory frameworks, and consumer education programmes designed to promote healthier food choices while addressing the realities of modern consumption patterns.

Contini et al. [31] explored consumer behaviour (intention and consumption) toward convenience-processed foods by applying the SEM technique with nine input constructs and two output constructs to test a model using a dataset of 426 Italian consumers. Four input constructs belonged to the group of belief factors, while the remaining five were classified as personal trait factors. Although convenience foods are not always ultra-processed, they often overlap with UPFs due to industrial processing, extended shelf life, and the inclusion of preservatives or artificial additives. The study reveals that six out of the ten hypotheses were supported, while the remaining three—related to value for money, taste, healthiness (from the belief factors group), and monetary resources (from the personal traits group)—were not supported. The hypothesis regarding the influence of intention on consumption was confirmed.

To examine changes in consumer behaviour, including fast-food consumption and online purchasing patterns resulting from the pandemic, Yan et al. [32] adopted the value–attitude–behaviour (VAB) model and designed two models: a first-order and a second-order constructs model to predict buying intention among 325 young people who live in Bangladesh and consume fast food. The first model analysed the relationships between seven input constructs—convenience, food quality, novelty seeking, subjective norms, and self-identification—and three output constructs: cognitive attitude, affective attitude, and buying intention. In the second-order model, utilitarian and hedonic values were considered higher-order reflective constructs. Utilitarian value was represented by convenience and food quality, while novelty seeking and subjective norms were interpreted as hedonic values. The study found that both higher-order constructs positively influenced cognitive and affective attitudes toward online fast-food purchasing. These attitudes, along with self-identity and subjective norms, were strongly and positively correlated with behavioural intention, whereas the role of food quality was found to be statistically insignificant.

Using the theory of consumption values, Arroyo [33] investigated how consumers rank food choice values using a dataset of 256 completed questionnaires from Mexican consumers. The questionnaire was designed to measure seven food choice values along with selected psychographic constructs, including safety and sustainability, weight control, convenience, basic sensory attributes, traditionalism, emotional value, novelty of functional food, self-control, self-assessment of diet quality, and health consciousness. For data analysis, in addition to classical statistical methods, the author introduced decision trees (DTs) as an innovative analytical tool to better understand consumer choices. The results empirically validated food choice value metrics and revealed the underlying motives influencing snack and beverage choices.

Calvo-Porral et al. [34] examined the relevance of nine factors influencing consumer acceptance of UPF products in Spain. This study proposed a measurement scale incorporating quality, time savings, price, effortless preparation, convenience, hedonism, marketing strategies, purchase intention, and satisfaction. The objective was to develop an evaluation framework for assessing UPF consumption and demand. Using a survey-based approach, the study collected a dataset from 478 completed questionnaires. The results from a confirmatory factor analysis (CFA) confirmed that the standardised factor loadings of the first-order constructs were valid and that the model fit was adequate. The study provided marketers and food companies with a validated instrument to measure the consumer acceptance and consumption of UPF products.

Norfalah et al. [35] developed a research model to examine the moderating effect of consumer value on UPF consumption in Malaysia using partial least squares structural equation modelling (PLS-SEM). The conceptual framework was based on five constructs: functional, emotional, conditional, and social values, along with UPF continuance consumption. An analysis of the dataset, which included responses from 286 university students, confirmed the research hypotheses that UPF continuance consumption directly depends on these constructs. Furthermore, AI-based recommendations were found to enhance both the emotional appeal and promotional effectiveness, potentially increasing UPF consumption.

Raj et al. [36] proposed and verified the effects of extrinsic (price, packaging, advertisement, brand, labelling, and word of mouth) and intrinsic (nutritional composition, taste, aroma, and texture) factors on the attitude and purchase intention towards UPFs using 375 responses from India. The conceptual model includes four key constructs and eight items per construct. The obtained results showed that the impact of extrinsic input construct on the output variables is statistically significant; however, the intrinsic factor impacts only attitude but is not significantly correlated with purchase intention.

Stamatelou et al. [37] developed a research model to determine the impact of knowledge and perceptions on consumption of the Greek population. The data were collected from 1126 adults through an online questionnaire with 16 questions for knowledge and perceptions assessment. To assess consumption, the relevant data were collected through a food frequency questionnaire. The analysis of the obtained results revealed that 40% of the participants have limited to no awareness regarding the term UPFs. An inverse dependence was observed between the knowledge and perception score and the overall consumption score of UPFs.

Yuan et al. [38] investigated the key factors influencing consumer knowledge, attitudes, and practices (KAPs) regarding preservatives among 515 pregnant women in China using regression analysis (RA) and SEM. In this cross-sectional study, the authors found that, while pregnant women had limited knowledge, they exhibited favourable attitudes and proactive practices toward harmful food components. The authors emphasised that improving knowledge, particularly concerning food safety, should be a focus of educational interventions.

Van der Merwe et al. [39] examined the relationship between consumers’ subjective and objective knowledge of healthy foods and various healthy lifestyle choices among 157 South African corporate employees. SEM confirmed a significant association between subjective knowledge and most healthy lifestyle choice categories but found no such link with objective knowledge. The findings of this cross-sectional study further emphasise the need to account for consumer demographic diversity when designing health education programmes to improve their effectiveness, particularly in socially and economically diverse societies.

Zheng at al. [40] analysed consumers’ willingness to purchase prepared dishes (PDs) and the moderating role of time pressure using SEM and 403 completed questionnaires of consumers in China. The results showed that consumers’ attitudes towards buying PDs, perceived behavioural control, and subjective norms have a significant positive effect on purchase intention. However, behavioural attitudes only served as a full mediator between functional value, emotional value, and purchase intentions.

Consumer willingness to purchase UPFs is influenced by multiple factors, including taste preferences, convenience, price sensitivity, and health awareness. While some consumers are drawn to UPFs due to their affordability and ease of access, others may actively avoid them due to concerns over nutritional quality and long-term health risks. The perception of UPFs as either beneficial or harmful is often shaped by advertising strategies, food labelling, and expert recommendations, all of which contribute to varying degrees of consumer trust and acceptance.

3.2. Comparison of Existing Models of User Attitudes Towards Industrially Formulated Foods

The studies presented in the preceding subsection are grounded in theoretical frameworks such as the health belief model, the theory of planned behaviour, and the knowledge–attitude–practice (KAP) model, which explore the psychological, social, and behavioural determinants of food consumption. Key factors in these models include health consciousness [33], knowledge of UPFs [37,38,39], social norms [31,32], and sustainability concerns [33], all of which influence consumer attitudes, willingness to purchase, and actual buying behaviour.

The TPB posits that an individual’s intention to perform a behaviour is influenced by their attitudes towards the behaviour, subjective norms, and perceived behavioural control. This framework has been applied to predict healthy eating behaviours, demonstrating that attitudes and intentions are mediated by goals and needs [31,32,36,40].

The HBM examines how personal beliefs about health conditions influence health-related behaviours. It suggests that individuals’ perceptions of susceptibility to illness, the severity of health conditions, benefits of preventive actions, and barriers to healthcare access determine their engagement in health-promoting behaviours [38].

The KAP model assesses the relationship between knowledge, attitudes, and practices within specific contexts. It has been utilised to understand how educational interventions can modify dietary behaviours by enhancing knowledge and shaping positive attitudes toward healthy eating [32,37,38].

Collectively, these theoretical models provide comprehensive insights into the multifaceted factors influencing dietary decisions, thereby informing the development of effective interventions to promote healthier food choices.

Most of the studies investigating consumer behaviour towards food products have utilised SEM [31,32,35,36,38,39,40] and confirmatory factor analysis (CFA) [33,34] due to their suitability for complex models with multiple latent variables and smaller sample sizes. In addition, some models have been developed using alternative approaches, such as ML algorithms like classification trees, for analysis to uncover hidden patterns in consumer data [33].

The key characteristics of the models addressing the factors influencing consumer behaviour towards UPFs are summarised in

Table 1. Comparison of models on UPFs perception: consumer attitudes, purchase intentions, and actual buying behaviour.

Reference	Study Design	Evaluation Factors (Number)	Statistically Significant Factors (Number)	Model Quality (s)
Contini et al., 2018 [31]	SEM	Value for money, taste, naturalness, healthiness, cooking skills, time pressure, monetary resources, social influence, market availability (9) → intention → consumption	Naturalness, cooking skills, time pressure, social influence, market availability (5)	R2: 0.74; 0.63
Yan et al., 2022 [32]	SEM	Convenience, food quality, novelty seeking, subjective norms, self-identification, utilitarian value, hedonic value (7) → cognitive attitude, affective attitude (2) → buying intention (1)	Convenience, novelty seeking, subjective norms, self-identification, utilitarian value, hedonic value (6)	R2: 0.85, 0.77; 0.79
Arroyo 2023 [33]	CFA, DTs	Safety and sustainability, weight control, convenience, basic sensory attributes, traditionalism, emotional value, novelty of functional food Self-control, self-assessment of diet quality, health consciousness (10)	Safety and sustainability, weight control, convenience, basic sensory attributes, traditionalism, emotional value, novelty of functional food Self-control, self-assessment of diet quality, health consciousness (10)	Accuracy: 0.78, 0.90
Calvo-Porral et al., 2024 [34]	CFA	Quality, time, price, effortless preparation, convenience, hedonism, marketing strategies (7) → satisfaction → purchase intention	Quality, time, price, effortless preparation, convenience, hedonism, marketing strategies (7)	RMSEA: 0.05
Norfarah et al., 2024 [35]	SEM	Functional, emotional, confidential, social values (4) → continuance consumption	Functional, emotional, confidential, social values (4)	R2: 0.50
Raj et al., 2024 [36]	SEM	Extrinsic factors, intrinsic factors (2) → attitude → purchase intention	Extrinsic factors, intrinsic factors—partial effect (2)	RMSEA: 0.05; 0.07
Stamatelou et al., 2024 [37]	Statistical methods	Knowledge and perceptions (1) → consumption	Knowledge and perceptions (1)	ST: 0.04
Yuan et al. 2024 [38]	RA, SEM	Knowledge, attitude, practice (3)	Knowledge, attitude, practice (3)	RMSEA: 0.06
Van der Merwe et al., 2024 [39]	SEM	Knowledge, income (2) → food and sleep, exercise and relaxation, dedicated efforts, not smoking	Knowledge, income (2)	RMSEA: 0.07
Zheng et al., 2024 [40]	SEM	Functional, emotional, confidential values (3) → attitude, perceived control → willingness to buy	Functional, emotional, confidential values (3)	RMSEA: 0.06; R2: 0.52; 0.52

Note: In the last column, R² represents the determination coefficient, RMSEA denotes the root mean square error of approximation, and ST indicates the standard error. The symbol “;” separates the coefficient value(s) for “Attitude(s)” from those for “Purchase Intention” and “Consumption.”

, providing an overview of the methodologies, theoretical foundations, and variables explored in this research domain.

The distribution of constructs in the above-mentioned models is as follows: purchase intention—5/10; attitude (including satisfaction)—5/10; knowledge (including cognitive attitude)—4/10; emotional value—3/10; consumption—3/10; food quality (including taste)—3/10; price (including value for money, income, and monetary resources)—3/10; convenience—3/10; functional value—2/10; time (including time pressure)—2/10; social influence (including subjective norms)—2/10; etc. The effectiveness of the models proposed in the literature, as measured using the coefficient of determination (R²), ranges from 50.0% [35] to 85.0% [32]. The root mean square error (RMSEA) values span between 0.05 and 0.08, with the number of latent variables varying from 2 to 11. The count of statistically significant factors fluctuates within this range.

Despite extensive research on the factors shaping consumer attitudes toward UPFs, there is still no widely accepted framework for assessing their nutritional impact and health consequences. Moreover, studies on UPF consumption patterns and their broader implications, particularly across diverse socio-economic and cultural contexts, remain limited. The existing research does not fully capture the evolving nature of dietary preferences, public health awareness, or data analysis methods. Therefore, advancing methodological approaches and conducting empirical investigations in this field can help bridge these gaps and generate valuable insights for public health policymakers and the food industry.

3.3. Main Factors Affecting Consumer Attitudes Towards Ultra-Processed Foods and Their Impact on Buying Decisions

The HBM [29] and the TPB [30] can be adapted and applied to understand consumer attitudes and decision-making regarding UPFs. Below is a brief overview of the key constructs in this model. These constructs shape consumer attitudes, which, in turn, influence purchase intentions and actual buying behaviour. The outlined framework offers a holistic perspective illustrating how individual factors (such as health awareness and product familiarity) and external social influences (such as subjective norms and environmental considerations) interact to drive UPF consumption.

Health consciousness reflects an individual’s awareness and concern for their well-being, shaping daily decisions, including dietary choices. It represents the motivation to adopt behaviours that promote health and mitigate risks. Prior research indicates that higher health consciousness significantly affects food purchasing behaviour, with higher health awareness leading to a preference for natural, minimally processed foods, and a reduced intake of UPFs [37,41].

Knowledge about UPFs refers to an individual’s understanding of the composition, processing methods, and health implications associated with these products. Knowledge can be both objective (factual information) and subjective (perceived understanding based on personal beliefs). Studies indicate that consumers with higher levels of nutritional knowledge are better equipped to make healthier food choices and are more likely to limit their consumption of UPFs [33,37,40].

Social norms encompass the shared beliefs, values, and behaviours within a community or social group that shape individual decision-making. In the context of UPF consumption, social norms can shape dietary habits through peer influence, family practices, cultural expectations, and media exposure [42]. Studies suggest that individuals tend to adopt eating behaviours aligned with their social circles, particularly under the influence of family, friends, and online communities [31,32,35].

Environmental concerns reflect an individual’s awareness of and motivation to address the environmental impact of their consumption choices. These concerns encompass resource depletion, greenhouse gas emissions, plastic pollution, and the ecological footprint of UPF production, packaging, and distribution. As sustainability gains global importance, consumers are increasingly considering the broader environmental consequences of their dietary habits [8]. Research suggests that environmentally conscious individuals are more likely to prefer foods with a lower ecological impact and tend to reduce UPF [18,33].

Attitude towards UPFs refers to an individual’s overall positive or negative evaluation of these products, encompassing both affective (emotions and feelings) and cognitive (knowledge and beliefs) components. These attitudes are influenced by various factors, including personal beliefs, health consciousness, social norms, marketing exposure, and environmental awareness [43]. Empirical studies indicate that negative attitudes towards UPFs are often driven by concerns about their nutritional quality, potential health risks, and environmental impact, whereas positive attitudes are typically associated with their convenience and affordability [32,36].

Willingness to purchase UPFs represents a behavioural intention and reflects the extent to which consumers are inclined to buy these products in the future. It is influenced by attitudes, perceived social norms, personal values, and product-related factors such as price, taste, and convenience [36]. Demographic variables such as age, income level, and education significantly moderate the relationship between attitudes and purchase intentions towards UPFs [40].

Actual buying behaviour towards UPFs refers to the observed purchasing patterns and consumption habits of individuals concerning these products. It represents the final stage in the decision-making process, influenced by prior attitudes, purchase intentions, and external factors such as marketing strategies, availability, and socio-economic status. Research indicates that, despite growing health awareness, UPFs continue to dominate food markets globally due to their convenience, aggressive advertising, and widespread availability in modern retail environments [32]. Price sensitivity, time constraints, and taste preferences often outweigh health-related concerns, leading to the frequent consumption of UPFs, especially among younger populations and urban dwellers [40]. Additionally, impulsive buying tendencies, influenced by promotional offers and attractive packaging, significantly affect the actual purchase of UPFs.

Based on the synthesis and comparison of existing models for consumer attitudes towards UPFs (Table 1) and an analysis of key factors influencing food-related decision-making, the following hypotheses are formulated:

H₁: 

There is a significant impact of health consciousness on consumer attitudes towards ultra-processed foods.

H₂: 

There is a significant impact of knowledge about UPFs on consumer attitudes towards ultra-processed foods.

H₃: 

There is a significant impact of social norms on consumer attitudes towards ultra-processed foods.

H₄: 

There is a significant impact of environmental concerns on consumer attitudes towards ultra-processed foods.

H₅: 

There is a significant impact of attitude towards ultra-processed foods on consumer willingness to purchase.

H₆: 

There is a significant impact of consumer willingness to purchase towards their actual buying behaviour.

H₇: 

Customer demographic characteristics have statistically significant mediating effects on consumer attitudes, willingness to purchase, or purchase decisions.

Note: the demographic characteristics include gender, age, educational level, and residence.

The proposed structural model (Figure 1) integrates psychological, social, and environmental constructs, reflecting the complex nature of consumer decision-making processes regarding food choices.

The proposed hypotheses help determine the impact of key factors on various aspects of consumer attitudes toward UPFs. In the next section, we present a holistic approach to analysing how these factors determine purchasing behaviour concerning attitudes toward UPF products.

4. Research Methodology

The formulated hypotheses aim to reveal the significance of the identified factors on various aspects of consumer attitudes towards UPFs. In the following section, this study adopts an integrated approach to evaluating how these factors influence consumer purchasing behaviour, focusing on the relationships between health consciousness, knowledge about UPFs, social norms, environmental concerns, and attitudes towards UPFs, as well as their impact on willingness to purchase and actual buying behaviour.

4.1. Questionnaire Design and Data Collection

The survey approach was selected as the primary method for gathering comprehensive data on consumer attitudes toward UPFs. An online questionnaire was developed, following established methodologies [31,32,33], and comprised sections on introduction, demographics, experience with UPFs, attitudes towards them, purchasing intentions, and their actual consumption. Each construct was measured using items adapted from previous studies to ensure both relevance and validity in capturing the determinants of UPF consumption.

Indicators for Question 9 (health consciousness) and Question 10 (knowledge) were adapted from the research of Yan et al. [32], Yuan et al. [38], and van der Merwe [39], respectively. The health consciousness and knowledge factors consist of three and two indicators, respectively. The two items for Question 11 (subjective norms) were sourced from the works of Contini et al. [31], Yan et al. [32], and Norfalah et al. [35]. Question 12, which measures environmental concerns, was obtained from the work of Arroyo [33]. The indicators for Question 13 (attitude) were adapted from the research of Yan et al. [32], Calvo-Porral et al. [34], Raj et al. [36], Yuan et al. [38], and Zheng et al. [40]. The indicators for buying intention (Question 14) and consumption (Question 15) came from Contini et al. [31], Yan et al. [32], Calvo-Porral et al. [34], Raj et al. [36], and Zheng et al. [40] and Contini et al. [31], Norfalah et al. [35], Stamatelou et al. [37], and Yuan et al. [38], respectively. The last four factors (environmental concerns, attitude, willingness to purchase, and actual buying behaviour) consist of three indicators each. To incorporate the participants’ opinions and suggestions, the last question (Question 16) was included based on recommendations from food marketing experts [44]. The research details and questionnaire link were disseminated through partner organisations via classic web and social media communications.

In the next section, our data models operationalise these constructs in a sequential framework, whereby health consciousness, UPF knowledge, subjective norms, and environmental concerns are posited to influence overall consumer attitude, willingness to purchase, and actual buying behaviour. Grounded in established theory and empirical evidence, these models help identify key determinants of UPF consumption and inform potential interventions to promote healthier dietary choices.

4.2. Questionnaire Measurements and Scales

Half of the survey questionnaire questions (8 out of 16) are structured as “multiple choice grid” questions, utilising a five-point Likert scale ranging from “strongly disagree” to “strongly agree”. An additional 30% of the questionnaire (5 out of 16) consists of “multiple choice” questions. Two questions require open-ended text responses, entered into text fields designated as “short answer” or “paragraph” in Google Forms. Finally, one question is formulated using a five-point linear scale.

4.3. Data Analysis Methods

The analytical approaches used to explore consumer attitudes towards ultra-processed foods (UPFs) can be broadly categorised into three groups: statistical methods, ML techniques, and MCDM approaches.

The first group, statistical methods, focuses on measuring, summarising, and visualising the key characteristics of multi-dimensional datasets. This includes techniques such as normality tests, t-tests, analysis of variance (ANOVA), chi-squared tests, and regression analysis. A prominent method within this category is PLS-SEM, which is well suited to analysing complex models with latent variables, particularly when working with smaller sample sizes.

The second group, ML methods, encompasses techniques like cluster analysis, predictive modelling, and sentiment analysis. These methods help uncover hidden patterns and relationships within data. For example, cluster analysis identifies groups of consumers with similar attitudes, while sentiment analysis evaluates subjective opinions from text data. Unlike traditional statistical approaches that emphasise hypothesis testing and causal inference, ML methods are data-driven, focusing on pattern recognition and predictive analytics based on historical data.

The third category, MCDM methods, offers a structured framework for evaluating complex decisions where multiple factors must be considered simultaneously. In the context of UPF research, MCDM techniques can be applied to assess consumer preferences, rank influencing factors, and support policy-making decisions related to public health and nutrition.

Statistical methods such as SEM are suitable for hypothesis testing and examining theoretical models, while ML techniques excel at identifying trends and making predictions from large datasets. Alternatively, MCDM approaches provide a comprehensive framework for analysing multiple factors simultaneously, offering multi-faceted insights into consumer behaviour and preferences.

The integration of these diverse analytical methods enhances UPF research by providing a holistic view, improving the accuracy of findings, enabling detailed consumer segmentation, supporting data-driven decision-making, and fostering more effective health policy recommendations.

5. Data Analysis

The proposed methodology (Section 4) was applied to address the research tasks.

5.1. Data Collection and Preprocessing

We disseminated the online survey link through institutional websites, email, and social media channels. The survey targeted online Bulgarian consumers, and participation was entirely voluntary. Created using Google Forms, the questionnaire consisted of 16 items designed to assess consumer perceptions of the variables explored in this study [44]. Data collection took place between 7 March 2024 and 26 May 2024, yielding a total of 290 completed responses.

The questionnaire and the dataset containing respondents’ answers are publicly available in an online repository [44]. The coding rules and processed dataset are also accessible online. Of the 16 questions, responses to 14 were systematically coded. Additionally, open-text responses (related to municipality and opinions) underwent qualitative processing. Preprocessing procedures such as tests for observation reliability and similarities were applied to ensure data accuracy, consistency, and readiness for analysis. To ensure data quality, we performed a duplicate check. Although no fully identical rows were found, partial duplicates were identified within responses related to the model constructs (Questions 9 to 16) for six participants (#2, #6, #7, #101, #206, and #274). As none of the entries were completely identical across all variables, all responses were retained for analysis.

5.2. Preliminary Statistical Analysis

To profile the survey participants, descriptive statistical analyses were conducted, including percentage distributions, descriptive statistics, and correlation analysis.

Table 2. Consumer profiles in the sample (n = 290).

Variables of the Sample		No. of Consumers	Percentage (%)
1. Gender	Male	68	23.4
	Female	222	76.6
2. Age	Under 20	229	41.0
	Between 21 and 30	67	23.1
	Between 31 and 40	33	11.4
	Between 41 and 50	37	12.8
	Over 50	34	11.7
3. Place of residence	City	196	67.6
	Town	88	30.3
	Village	6	2.1
4. Municipality		-	-
5. Monthly income per household member	Less than BGN 1438	137	47.2
	More than BGN 1438	153	52.8
6. Education	High school	168	57.9
	Bachelor	59	20.3
	Master	58	19.7
	PhD	6	2.1
7. Share of UPFs in your daily menu	I do not consume (0%)	26	9.0
	Less than 25%	149	51.4
	Between 25% and 50%	97	33.4
	Between 50% and 75%	15	5.2
	Over 75%	3	1.0
8. How are the UPFs you consume divided by product group (I do not consume (0%), less than 25%, between 25% and 50%, between 50% and 75%, over 75%)	Snacks	53, 142, 70, 19, 6	18.3, 49.0, 24.1, 6.6, 2.1
	Frozen meals	107, 136, 34, 13, 0	36.9, 46.9, 11.7, 4.5, 0.0
	Fast food	68, 140, 54, 23, 5	23.4, 48.3, 18.6, 7.9, 1.7
	Packaged breads and pasta	52, 136, 54, 32, 16	17.9, 46.9, 18.6, 11.0, 5.5
	Processed meat	46, 135, 61, 39, 9	15.9, 46.6, 21.0, 13.4, 3.1
	Sweetened cereals	125, 113, 32, 14, 6	43.1, 39.0, 11.0, 4.8, 2.1
	Sweetened dairy products	82, 135, 46, 20, 7	28.3, 46.6, 15.9, 6.9, 2.4
	Instant soups and pasta	177, 76, 24, 11, 2	61.0, 26.2, 8.3, 3.8, 0.7

summarises the demographic characteristics of the survey participants. The majority were female, representing 77% of the total sample (Question 1). Participants under the age of 30 comprised 64% of the respondents (Question 2). Furthermore, the survey predominantly included individuals from urban areas, accounting for 97.6% of the total responses (Question 3).

The majority of survey participants were from Plovdiv Province, accounting for 71% of the total responses. Respondents from Haskovo Province made up 7%, while those from Pazardzhik Province represented 6% of the sample (Question 4). Overall, the survey predominantly covered the South-Central region, which included 88% of the participants, followed by the Southeastern and Southwestern regions with 7.9% and 2.4%, respectively.

Our respondents were nearly equally divided into two groups based on their family income: 47% had an income below the necessary net monthly income for the maintenance of a working person, while 53% had an income above this threshold (Question 5). In terms of educational background, 42.1% of the respondents had attained education beyond secondary school, while 57.9% held a secondary school degree (Question 6).

Most respondents consumed UPFs in small amounts, with the majority keeping their intake below 25% of their daily diet. Only a small portion of respondents (6.2%) had diets in which UPFs constituted the majority of their intake (Question 7). In contrast, the findings from a study analysing data from the National Health and Nutrition Examination Survey (NHANES) between 1999 and 2018 showed that the estimated percentage of total energy intake from UPFs among U.S. youths increased to 67.0% [45]. One possible explanation is that the respondents in this sample belonged to Generation Z and Generation Alpha classifications, while our sample was more diversified. The obtained weighted average share of UPFs daily consumption (34.5%) aligns with the findings from a study examining household availability of UPFs across nineteen European countries. This study reported that the average household availability of UPFs ranged from 10.2% in Portugal and 13.4% in Italy to 46.2% in Germany and 50.4% in the UK. This suggests that our country has a middle prevalence of UPF consumption in comparison with other European countries [46].

Processed meat, packaged breads and pasta, and snacks are among the most frequently consumed UPFs. In contrast, instant soups and pasta, sweetened dairy products, sweetened cereals, and frozen meals have the highest non-consumption rates (Question 8). These findings partially align with those from a study on convenience food consumption in Germany [47], which identified sweetened yogurt, pizza, and chips as among the most widely consumed products.

5.3. Clustering

To identify groups of consumers with similar characteristics and determine the variables influencing their attitudes, an unsupervised machine learning approach—cluster analysis—was performed using the k-means algorithm (Figure 2 and

Table 3. Average values by cluster and the absolute differences between clusters shown by construct indicator.

	HC1	HC2	HC3	UPFK1	UPFK2	SN1	SN2
Cluster 1	4.119	4.552	3.858	4.045	4.455	2.828	3.858
Cluster 2	2.628	3.660	2.840	3.256	3.654	2.160	2.449
Difference	1.491	0.892	1.018	0.788	0.801	0.668	1.409
	EC1	EC2	EC3	ATT1	ATT2	ATT3	WP1
Cluster 1	4.194	3.672	3.985	4.172	4.463	4.097	3.090
Cluster 2	3.385	2.577	3.173	3.128	3.571	3.128	2.128
Difference	0.809	1.095	0.812	1.043	0.892	0.969	0.961
	WP2	WP3	ABB1	ABB2	ABB3
Cluster 1	4.187	3.694	3.097	3.381	3.970
Cluster 2	2.994	2.596	2.212	3.237	2.667
Difference	1.193	1.098	0.885	0.143	1.303

). The optimal number of clusters was determined using the Elbow and Silhouette methods, with both indicating that two clusters provided the best solution. These clusters consisted of 134 and 156 respondents, respectively. As shown in Figure 2, the clusters are well defined when k = 2, demonstrating that the k-means algorithm effectively differentiates consumer groups with comparable attitudes towards UPFs.

The first cluster (Cluster 1) comprised 134 customers, demonstrating more awareness of the consumption of UPFs. They exhibited higher ratings for their health concerns (Question 9, HC), better knowledge (Question 10, UPFK), less social pressure (Question 11, SN), environmental concerns (Question 12, EC), attitudes towards UPFs (Question 13, ATT), buying intention (Question 14, WP), and actual buying behaviour (Question 15, ABB) (Table 3). Among the indicators, knowledge (Question 10, UPFK), attitudes (Question 13, ATT), and health consciousness (Question 9, HC) exerted the strongest influence on overall UPF perception. Meanwhile, the second cluster (unconcerned consumers) reflected fewer concerns about social norms (Question 11, SN) and less knowledge about UPFs (Question 10, UPFK), with a minimal intention to buy (Question 14, WP) and consume these harmful foods (Question 15, ABB). Table 3 illustrates the average values of the indicators for both clusters, along with the differences between these estimates.

5.4. Sentiment Analysis

Following preprocessing and filtering, 113 responses remained, reflecting user perceptions about this food category. They were analysed using sentiment analysis, a machine learning technique, yielding the following results: 30 negative responses (average score: −0.172), 28 neutral responses, and 55 positive responses (average score: 0.241). Overall, respondents generally did not support the spread of UPFs despite their convenience and lower cost, particularly for busy individuals or those on a budget. The sentiment analysis was performed by computing a polarity value for each text entry and categorising it as positive, negative, or neutral, based on its numerical value, using Python 3.12 and the TextBlob 0.19.0 library.

The detailed analysis showed that a significant portion of the group wished to reduce or avoid UPF consumption, citing factors such as newly acquired nutrition knowledge, personal health improvements, and a desire to make better dietary choices. However, convenience and lower costs remained major barriers for others, particularly those in busy or low-income households, as UPFs are often more accessible and affordable. The call for stronger regulations is evident among a noteworthy number of respondents, who suggested policies ranging from better product labelling to outright bans, emphasising the need for government intervention to protect public health. While a significant number of respondents reported a shift in their attitudes toward UPFs, induced via increased awareness and education, others remain indifferent or unchanged in their stance. Throughout, the tension between personal choice and systemic issues was evident, as many argued that increased public awareness, stricter regulation, and the availability of healthier food alternatives could help curb UPF consumption.

5.5. SEM Model of Customer Attitudes and Purchase Behaviour Towards UPFs

Based on a review of prior studies (see Section 3), there is a lack of consensus regarding the definition of inputs and outputs for evaluating consumer perceptions and attitudes towards industrially formulated food products. To address this issue, we applied the PLS-SEM method using SmartPLS (version 3.2.9) [48]. Additionally, a standard five-step procedure was followed for constructing the PLS-SEM model.

Step 1. Formulate hypotheses.

Seven research hypotheses were formulated at the end of Section 3, focusing on the relationships between the input and output variables.

Step 2. Identify indicators for latent variables.

Six constructs were measured by 19 indicator variables from the survey questionnaire. The measurement model includes 10 input indicators: health consciousness (HC): HC1, HC2, and HC3; knowledge about UPFs (UPFK): UPFK1 and UPFK2; subjective norms (SNs): SN1 and SN2; and environmental concerns (ECs): EC1, EC2, and EC3.

It also comprised nine output indicators: attitude towards UPFs (ATT): ATT1, ATT2, and ATT3; willingness to purchase (WP): WP1, WP2, and WP3; and actual buying behaviour (ABB): ABB1, ABB2, and ABB3.

These relationships are illustrated in Figure 3.

Step 3. Conduct numerical modelling.

We ran the PLS algorithm to estimate model parameters and evaluate the measurement and structural models.

Step 4. Assess the model’s suitability.

If the model fits the data, proceed to Step 5. Otherwise, return to Step 3 to refine it.

The initial assessment of outer loadings showed that most indicators fit well, except ABB1 and ABB2, which fell below the acceptable threshold. Consequently, both were removed, bringing all values within acceptable limits and allowing us to proceed with model reliability and validity verification (Step 4).

In the first phase of the validity assessment, the measurement model was evaluated first, followed by the structural model. The former ensures construct validity and reliability by assessing construct reliability, indicator reliability, convergent validity, and discriminant validity, while the latter verifies the significance of the proposed relationships.

Factor loadings represent how strongly each item correlates with its designated principal component. Higher absolute values indicate a stronger relationship between an item and its underlying factor. In our study, all items have loadings above the recommended threshold of 0.5 [49,50]. The resulting measurement model and corresponding factor loadings are presented in

Table 4. Factor loadings for indicators.

Indicator Variable	Factor Loading	Indicator Variable	Factor Loading	Indicator Variable	Factor Loading
HC1	0.834	SN2	0.933	ATT3	0.893
HC2	0.859	EC1	0.864	WP1	0.612
HC3	0.780	EC2	0.759	WP2	0.851
UPFK1	0.811	EC3	0.790	WP3	0.873
UPFK2	0.930	ATT1	0.852	ABB3	1.000
SN1	0.790	ATT2	0.842

.

To assess multicollinearity among indicators, the study employed the variance inflation factor (VIF). A VIF value below five suggests acceptable levels of multicollinearity.

Table 5. Construct reliability (Cronbach’s alpha, DG rho, and CR), convergent validity (AVE), and multicollinearity (VIF).

Factor	Cronbach’sAlpha	DG rho	CR	AVE	VIF
Health consciences	0.770 *	0.799 *	0.864 *	0.680 *	1.781 *
Knowledge about UPFs	0.700 *	0.804 *	0.864 *	0.761 *	1.484 *
Subjective norms	0.681 *	0.817 *	0.855 *	0.747 *	1.356 *
Environmental concerns	0.731 *	0.751 *	0.847 *	0.649 *	1.484 *
Attitude towards UPFs	0.828 *	0.831 *	0.897 *	0.744 *	1.000 *
Willingness to purchase	0.680 *	0.708 *	0.827 *	0.620 *	1.000 *
Actual buying behaviour	1.000	1.000 *	1. 000 *	1.000 *

Symbol “*” means Cronbach’s alpha > 0.6, DG rho—Dillon–Goldstein’s rho > 0.7, CR—composite reliability > 0.6, AVE—average variance extracted > 0.5, and VIF—variance inflation factors < 5.

presents the VIF values for each indicator, all of which are within the acceptable range [51].

In the presented table, most of the constructs demonstrate acceptable to high internal consistency, emphasising the reliability of the measurement scales. However, subjective norms (α = 0.681) and willingness to purchase (α = 0.680) show relatively lower reliability, suggesting that minor improvements to these measurement items may be beneficial. Nonetheless, the remainder of the constructs meet the recommended criteria, validation the overall reliability of the model.

There are two main approaches to assessing construct reliability (i.e., repeatability): Dillon–Goldstein’s rho (also known as rho_A in SmartPLS) and composite reliability (CR). For robust reliability, both DG rho and CR should be ideally above 0.7 [50]. In our study, the DG rho values ranged from 0.708 to 0.831, while the CR values spanned from 0.827 to 0.897 (see Table 5), indicating that the constructs exhibit acceptable reliability.

Construct validity was evaluated through assessments of both convergent and discriminant validity.

Convergent validity, which measures the consistency among different indicators of the same concept, was examined using the average variance extracted (AVE); a value of 0.5 or higher is generally considered satisfactory [51]. All constructs surpassed this threshold (Table 5), confirming good convergent validity.

Discriminant validity, or the degree to which constructs are distinct from one another, was assessed using the Fornell–Larcker criterion, which requires that the square root of the AVE for each construct exceeds its correlations with other constructs.

As detailed in

Table 6. Discriminant validity—Fornell and Larker criterion.

Factor	Actual Buying Behaviour	Attitude Towards UPFs	Environmental Concerns	Health Consciousness	Subjective Norms	Knowledge About UPFs	Willingness to Purchase
Actual buying behaviour	1
Attitude towards UPFs	0.509	0.863
Environmental concerns	0.540	0.662	0.806
Health consciousness	0.522	0.490	0.509	0.825
Subjective norms	0.512	0.427	0.429	0.456	0.864
Knowledge about UPFs	0.372	0.479	0.409	0.549	0.323	0.872
Willingness to purchase	0.632	0.529	0.507	0.477	0.454	0.281	0.788

Note: italics represent the square root of AVE.

, this condition was met for all constructs, providing strong evidence of discriminant validity.

The cross-loadings analysis examines whether each item is more strongly associated with its intended construct than with any other in the model. In this study, the findings (see

Table 7. Discriminant validity—cross-loadings.

Indicator Variable	Actual Buying Behaviour	Attitude Towards UPFs	Environmental Concerns	Health Consciousness	Subjective Norms	Knowledge About UPFs	Willingness to Purchase
ABB3	1.000	0.509	0.540	0.522	0.512	0.372	0.632
ATT1	0.417	0.852	0.555	0.390	0.413	0.390	0.490
ATT2	0.399	0.842	0.543	0.444	0.313	0.425	0.380
ATT3	0.496	0.893	0.613	0.437	0.377	0.425	0.495
EC1	0.385	0.624	0.864	0.453	0.305	0.476	0.328
EC2	0.567	0.490	0.759	0.394	0.492	0.194	0.600
EC3	0.371	0.468	0.790	0.377	0.254	0.283	0.322
HC1	0.516	0.403	0.463	0.834	0.435	0.403	0.507
HC2	0.333	0.478	0.449	0.859	0.261	0.555	0.289
HC3	0.482	0.301	0.327	0.780	0.491	0.369	0.417
SN1	0.301	0.260	0.276	0.290	0.790	0.198	0.291
SN2	0.538	0.444	0.438	0.467	0.933	0.335	0.463
UPFsK1	0.345	0.311	0.293	0.469	0.297	0.811	0.254
UPFsK2	0.318	0.494	0.405	0.496	0.279	0.930	0.245
WP1	0.503	0.247	0.281	0.237	0.231	0.062	0.612
WP2	0.472	0.516	0.473	0.489	0.387	0.350	0.851
WP3	0.526	0.457	0.424	0.378	0.432	0.221	0.873

) show that all items load more highly on their designated constructs (italicised) than on alternative constructs. Therefore, the examination of cross-loadings supports the confirmation of discriminant validity.

The heterotrait–monotrait (HTMT) ratio is used to evaluate the correlations between constructs to confirm discriminant validity. While literature thresholds generally range from 0.85 to 0.9, the results from this study (see

Table 8. Discriminant validity—HTMT.

Factor	Actual Buying Behaviour	Attitude Towards UPFs	Environmental Concerns	Health Consciousness	Subjective Norms	Knowledge About UPFs	Willingness to Purchase
Actual buying behaviour
Attitude towards UPFs	0.557
Environmental concerns	0.640	0.840
Health consciousness	0.612	0.597	0.663
Subjective norms	0.584	0.537	0.590	0.631
Knowledge about UPFs	0.452	0.603	0.536	0.734	0.449
Willingness to purchase	0.777	0.692	0.728	0.668	0.628	0.397

) indicate that all HTMT ratios are below the upper limit of 0.9 and are statistically significant.

The p-values indicate that both knowledge and subjective norms have a moderate impact on attitudes toward UPFs, while environmental concerns exert a significant effect. In addition, attitudes toward UPFs strongly influence willingness to purchase, and willingness to purchase, in turn, has a very high impact on actual buying behaviour. All p-values are below 1%, with the exception of the health consciousness effect on attitude toward UPFs, which exceeds 20% (see Figure 4 and

Table 9. The path coefficients of the relationship between latent variables.

Hypothesis	β	Sample Mean	SD	t Statistics	p-Values	R2	Q2
Attitude towards UPFs → willingness to purchase	0.529	0.530	0.048	11.029	0.000	0.280	0.169
Environmental concerns → attitude towards UPFs	0.491	0.485	0.063	7.823	0.000	0.508	0.365
Health consciousness → attitude towards UPFs	0.079	0.082	0.064	1.230	0.219
Subjective norms → attitude towards UPFs	0.117	0.120	0.051	2.306	0.021
Knowledge → attitude towards UPFs	0.197	0.203	0.054	3.624	0.000
Willingness to purchase → actual buying behaviour	0.632	0.631	0.044	14.511	0.000	0.399	0.390

). These results generally support our hypotheses—with the exception of H₁—and are in line with most previous research. Notably, all predictor variables have positive regression coefficients.

Step 5. Interpret the obtained results.

One possible explanation for the rejection of the first hypothesis (H₁) (β = 0.079 and p > 0.05) linking health consciousness to attitudes towards UPFs is that the majority of our respondents were young individuals. Younger consumers may prioritise other factors—such as taste, convenience, or social influence—over health concerns when evaluating food choices. This demographic might also perceive the health risks associated with UPFs differently than older populations, leading to a weaker relationship between health consciousness and their attitudes towards these foods. Another possible explanation is that the conceptualisation of health consciousness might not have fully captured the specific attitudes toward UPFs. Respondents who are generally health-conscious may still perceive UPFs as acceptable if they believe these foods offer convenience or align with other lifestyle priorities. This could be due to a gap between general health awareness and the nuanced understanding of UPF-related risks, thereby weakening the hypothesised link between health consciousness and attitudes toward UPFs.

The support for hypothesis H₂ (β = 0.197 and p < 0.001) linking knowledge to attitudes towards UPFs can be explained by the fact that greater knowledge can reduce uncertainty and enhance risk perception, prompting consumers to reassess their dietary choices. As individuals become more aware of the ingredients, processing methods, and long-term consequences of consuming UPFs, they are likely to adopt attitudes that favour healthier alternatives, thereby reinforcing the significant influence of knowledge on consumer attitudes. Additionally, increased knowledge often correlates with improved nutrition literacy and access to reliable information sources. This empowers consumers to critically evaluate UPF marketing claims and make informed decisions, further solidifying the positive relationship between knowledge and a more cautious, health-oriented attitude toward UPFs.

Subjective norms influence attitudes towards UPFs (H₃) (β = 0.117 and p < 0.05) by shaping individuals’ perceptions of what is considered acceptable or desirable within their social circles. When family, friends, or influential social groups express concerns about UPF consumption or promote healthier eating habits, individuals are more likely to adopt a similar attitude, aligning their personal beliefs with those of their peers. Additionally, social pressure and the desire for social conformity play crucial roles. People often adjust their dietary choices to meet the expectations of significant others, thereby reinforcing attitudes that reflect the prevailing social sentiment. This dynamic, rooted in social learning and normative influence, strengthens the relationship between subjective norms and a more cautious, health-conscious stance towards UPFs. However, with a beta of 0.117, this suggests a relatively weak but still significant influence of subjective norms on attitudes towards UPFs, indicating that while social pressures play a role, the impact is not as strong as other factors in shaping consumer attitudes.

Hypothesis H₄, linking environmental concerns to attitudes towards UPFs, is supported (β = 0.491 and p < 0.001)) due to growing awareness among consumers about the environmental impact of food production, particularly that of UPFs. Individuals who are conscious of the ecological consequences—such as carbon footprints, packaging waste, and resource depletion—are more likely to adopt an attitude that opposes the consumption of products associated with significant environmental harm. Increasingly, many consumers who value sustainability and care about the planet’s future are motivated to choose plant-based, minimally processed foods over UPFs, strengthening the link between environmental concerns and attitudes toward ultra-processed foods. H₄ is supported because consumers with strong environmental concerns tend to be more aware of the ecological impacts associated with ultra-processed food production, such as excessive packaging waste, high energy consumption, and carbon emissions. This heightened awareness leads them to form more critical attitudes toward UPFs, aligning with broader sustainable consumption practices and reinforcing the negative perception of UPFs within environmentally conscious social circles.

H₅ is supported (β = 0.529 and p < 0.001) because individuals with a positive attitude towards UPFs often associate them with greater convenience, taste, or perceived value, which further strengthens their willingness to purchase such products. Additionally, a positive attitude can influence consumer behaviour by creating a sense of trust and satisfaction, making the prospect of purchasing UPFs more appealing, especially when marketing messages align with their personal preferences. H₅ is supported because a positive attitude toward UPFs makes these products appear more attractive, reducing perceived risks and enhancing consumer confidence in their quality. Additionally, favourable attitudes often foster brand loyalty, leading to repeated purchase behaviour, while effective marketing and social influences further reinforce this positive perception. Moreover, when consumers associate UPFs with convenience and desirable taste attributes, their willingness to purchase is further increased.

H₆ is supported (β = 0.632 and p < 0.001) because a high willingness to purchase typically translates into actual buying behaviour, indicating that, when consumers express strong intentions, they are likely to follow through with their purchases. Additionally, this consistency suggests that willingness to purchase is a reliable predictor of market behaviour, reflecting stable consumer decision-making processes that drive the transition from intention to action.

Our testing partially supports H₇ regarding education level, unlike the rest of the moderating sub-hypotheses. For the paths between attitude and willingness to purchase (β = 0.085 and p = 0.113), and willingness to purchase and actual buying behaviour (β = –0.075 and p = 0.116), individuals who are more educated are less reliant on independent variables in comparison to their counterparts. This may be attributed to the enhanced critical thinking skills fostered via higher education, which enable these consumers to more effectively assess information and evaluate the credibility of claims. As a result, educated individuals tend to be more sceptical about marketing messages and endorsements related to UPFs, placing less emphasis on such factors when forming their attitudes.

The R² values presented in Table 9 are 0.508 for consumer attitude towards UPFs, 0.280 for buying intention, and 0.399 for consumption. This suggests that approximately 50.8% of the variance in consumer attitude, 28.0% of the variance in buying intention, and 39.9% of the variance in consumption are explained by their respective predictor variables. Specifically, consumer attitude is influenced by knowledge, subjective norms, and environmental concerns, buying intention is driven by consumer attitude, and consumption is determined by willingness to purchase. According to standard interpretative thresholds, the explanatory power of the model for consumer attitudes is moderate to strong, while the models for buying intention and consumption exhibit moderate explanatory power. Additionally, the Q² values (0.365, 0.169, and 0.390 for consumer attitudes, buying intention, and consumption) confirm the predictive relevance of the SEM model, as all values are above zero. This indicates the model’s ability to generate reliable out-of-sample predictions.

5.6. Other Models of Customer Attitudes Towards UPFs

To explore the relationships between the input and output constructs, we applied four supervised ML algorithms as detailed in

Table 10. Results of using supervised ML algorithms to predict user attitude towards UPFs.

ML Method	MSE	RMSE	MAE	R2
Decision tree	0.058	0.240	0.108	0.926
SVM	0.143	0.378	0.239	0.816
Random forest	0.027	0.164	0.091	0.965
AdaBoost	0.028	0.167	0.064	0.964

. The mean square error (MSE) measures the average squared difference between predicted and actual values, while the mean absolute error (MAE) reflects the average of the absolute differences. In addition, the root mean square error (RMSE) is defined as the square root of the MSE, offering an error metric in the same units as the output variable. AdaBoost consistently achieved superior performance across all evaluation metrics, followed by the random forest and decision tree methods.

While the SEM approach typically yields lower R² values (from 0.280 to 0.508) compared to ML techniques, its main strength lies in the clear interpretability of the results. In contrast, ML models often achieve higher accuracy (ranging from 0.816 to 0.965) but tend to produce predictions that are less transparent and more difficult to explain. Ultimately, the choice between these methods depends on the specific objectives and requirements of the analysis.

Applying the coefficients from the SEM, we can use well-established MCDM methods, such as the Visekriterijumska Optimizacija I Kompromisno Resenje (VIKOR), to evaluate alternatives by calculating the distance from the ideal solution to identify the best compromise solution. In this case, the obtained complex composite index assesses consumer satisfaction towards UPF alternatives. Another distance-based MCDM method is the combined compromise solution (CoCoSo) method. CoCoSo integrates both additive and multiplicative aggregation approaches to rank alternatives based on their weighted criteria values and the distances from ideal solutions, providing a compromise ranking of compared UPF options.

6. Conclusions

Ultra-processed foods have transformed the modern food industry, influencing consumer choices and dietary habits. This study contributes to the existing body of knowledge on consumer attitudes towards highly processed foods by integrating key behavioural and demographic factors into a variety of data science models. The findings highlight that health consciousness alone does not significantly impact consumer attitudes (H₁), challenging conventional theories that assume health awareness directly influences food choices. Instead, UPF knowledge, social norms, and environmental concerns (H₂, H₃, and H₄) emerge as strong determinants, reinforcing the role of external influences and informational awareness on consumer perceptions. Furthermore, the validated SEM model demonstrates the link between attitudes, willingness to purchase, and actual buying behaviour (H₅, H₆), aligning with established behavioural intention models. Additionally, the study underscores the role of demographic characteristics (H₇), particularly education on food-related decision-making. These insights enrich the theoretical understanding of how cognitive, social, and contextual factors interact in shaping consumer behaviour towards UPFs.

From a practical perspective, the study offers recommendations for policymakers, health experts, and food industry stakeholders to promote healthier and more sustainable food choices. The identification of two distinct consumer clusters suggests the need for tailored interventions—educational campaigns should focus on increasing UPF awareness among less-informed consumers while reinforcing positive behavioural patterns among those already concerned about health and environmental impact. Since social norms significantly influence attitudes, leveraging peer-driven marketing strategies and public health campaigns that emphasise community engagement can enhance their effectiveness. Moreover, the findings highlight the role of demographic characteristics, emphasising the need for targeted communication strategies that address different consumer segments based on education actions. Food manufacturers can also utilise these insights to reformulate products, ensuring transparency in labelling and incorporating sustainable production practices that align with consumer preferences. Lastly, educational institutions can play a pivotal role by integrating food literacy programmes that equip consumers with the knowledge needed to make informed dietary choices.

This study involved several limitations. First, the sample size and its characteristics may not have fully captured the diversity of demographic perspectives. Second, consumer perceptions are complex, influenced by a range of psychological, social, and cultural factors that were not fully addressed in our analysis. Third, the focus was solely on individual consumer attitudes, omitting valuable insights from marketing experts and business representatives, which may result in an incomplete picture of the broader UPF area.

Future research should aim to overcome these limitations by broadening the participant pool to include additional demographics, such as older generations, and by investigating both direct and indirect relationships between variables using advanced SEM techniques. Comparative studies across different countries, with particular attention to the moderating effects of socio-economic factors like income and region, would also be beneficial. Furthermore, exploring the evolution of consumer behaviour in a post-pandemic environment and applying fuzzy MCDM methods to understand the multi-attribute interdependencies influencing consumer satisfaction in UPF marketing are recommended to ensure that future strategies foster ethical, socially responsible, and inclusive digital transformations.