The Forecasting Model of the Impact of Shopping Centres in Urban Areas on the Generation of Traffic Demand

Abstract

The increase in traffic caused by new development affects the change in traffic conditions on the surrounding roads, and shopping centres are significant traffic generators. The development of local travel generation rates and their characteristics for individual land uses from the aspect of traffic demand is a reliable way to plan traffic in order to come up with preventive solutions to traffic problems, that is, prevention of possible negative consequences on traffic conditions in the street network occurring due to the construction of shopping centres. One of the main aims of this paper is to develop a model for objective assessment of the generated traffic demand for significant changes in land use, such as the construction of shopping centres in medium-sized towns. All these would be steps in the right direction for the promotion of reliable traffic planning and adoption of TIA for every new development before a decision regarding the change in land purpose has been made. This kind of process still has not been established systematically in either Bosnia and Herzegovina and the Republic of Serbia, or in surrounding countries. This paper focuses on the formulation of a model for determining the volume of traffic generated by shopping centres in medium-sized towns in two countries of the Southeast Europe region. The survey was conducted in eight different locations (cities) where there are shopping centres with common facilities. The analysis showed that the number of visitors and vehicles attracted by the shopping centre zone can be determined by a model based on a linear regression analysis. The analysis included exploring several different factors of trip generation in shopping centres, including the relationship between trip generation and combinations of several independent variables. The verification of the model was conducted in real conditions of the traffic flow generated by a shopping centre which was not the analysis subject when forming the forecasting model. In this way, the validity of the proposed model is credibly assessed. The developed model can be applied in the procedures of planning the construction of shopping centres in medium-sized cities in the Republic of Serbia and Bosnia and Herzegovina, and wider, in the region of Southeast Europe, in order to estimate the volume of generated traffic demand, that is, its impact on the conditions of traffic on the surrounding traffic network.

1. Introduction

In the process of urban planning, it is necessary to define the purpose of the land and, in accordance with this, the traffic demand, which is very rarely carried out in the practice of the countries of the region of Southeast Europe. Most often, this step is skipped or analyses and forecasts are made during the implementation and realization of previously adopted urban plans, that is, when conforming to the designated purpose of individual locations. The sequence of steps should be different, i.e., the analysis of the generated traffic demand should precede the definition of the purpose of the land, that is, the type, size and content of the facilities at a location. It is known that any change in land use in urban areas and construction causes changes in the modal split, that is, traffic demand. Changes in traffic demand parameters, that is, modal split, size and characteristics of the traffic flow, can be the cause of changes in traffic conditions and the level of service on the street network in the vicinity of newly built attractions and travel generators. The modal split and the temporal and spatial distribution of newly created trips to the street network depend on the location, size and content of the newly built attraction, the characteristics of the surrounding transport network, and also local influences. The existence of mutual links between land use and traffic, in such a way that each land use generates a certain number of movements, results in the need to establish adequate transport. This link shows that differences in the purpose of surfaces cause differences in the size, composition, and spatial and temporal distribution of travel. The level at which an urban environment will function largely depends on the setting of the appropriate transport system. Land use and transport are interdependent, so it is very important to understand the relationship between land use and transport in order to plan a sustainable and safe transport system [1]. The establishment of any new development generates additional trips that may have negative effects on the existing traffic network. In order to assess the impact of newly created traffic on the transport network and determine reasonable solutions, a traffic impact assessment study is performed [2].

In many countries, based on the results of the research, models have been defined on the basis of which it is possible to predict the characteristics of newly created trips in order to maintain or improve the existing traffic conditions through changes in the street network. Namely, each new attraction that occurs during land use change affects the generation of additional trips that can have negative effects on the existing transport network. In order to assess the impact of newly created traffic on the transport network and determine reasonable solutions, it is necessary to assess the impact on traffic demand before changing the land use. The travel matrix, modal split and distribution of traffic flows to the network, in addition to land use, also depend on local characteristics and impacts, which indicates the need for research in this area. Traffic impact assessments of newly built travel generators are defined in many countries by legislative acts as obligations of investors for newly built facilities and land use changes.

In addition to many developed countries having elaborate models and routes of their own to estimate the generation of travel, there are those that do not. In our region, no legality has been established, i.e., no models have been defined on the basis of which impact assessments of newly built travel generators could be performed due to land use change. Currently, there are no methodological guidelines in the Republic of Serbia and Bosnia and Herzegovina for assessing the volume of traffic generated by the construction of various facilities. Many countries, in the absence of their own models, most often use the travel rates from the manual to generate travel, which is published and regularly updated by the American Institute of Transportation Engineers (ITE). In addition to the ITE, there are guidelines for creating trips available in the literature dealing with this issue. However, if all these developed models and guidelines were directly applied in our area, which was determined by some different local characteristics, it would lead to a reduction in the accuracy of the assessment in the mildest form. Each region has its own local characteristics that are reflected in different land use, different degrees of motorization, modes of transport, socioeconomic characteristics, different existing transport systems and in general, different ways of life.

Trip generation hubs are places that concentrate on one or more activities, attract and produce high traffic demand and therefore need to be specifically studied. Travel that is attracted and generated for any mode of operation has an impact on traffic throughout the region where the hub is located and must be quantified when located in an urban area [3]. However, traffic impact assessment (TIA) analysis is not required for every construction or land use change because there are facilities that do not generate significant traffic, i.e., do not affect the change in traffic conditions. It is accepted practice that a comprehensive traffic impact assessment should be conducted whenever construction is expected to increase by 100 or more inbound or outbound trips during peak hours [4].

The generated traffic demand must be quantified, and shopping centres are singled out as generators, which attract and generate high traffic demand, mainly for the purpose of shopping and leisure. Depending on the size and attractiveness, malls in peak hours can also attract several hundred trips that as such create a significant additional load on the surrounding street network, which can cause a deterioration in the level of service. The number of new shopping centres that have appeared in the last 20 years sets them apart and puts them as one of the most important generators of new traffic demand. However, in many cities of the region, in various locations, shopping centres were built, mostly without any analysis and assessment of the impact of the newly created traffic on the level of service of the surrounding roads. For this reason, on the street network in the vicinity of some shopping centres, there has been a drastic deterioration in traffic conditions, which are manifested by congestion of the street network, time losses at intersections, increased emission of pollutants, etc.

All the above mentioned are the reasons that indicate the need to define a model for estimating traffic demand generated by shopping centres, in accordance with the local conditions prevailing in our region. This model would be used in the process of traffic planning, which is essential for the planning of transport infrastructure, traffic and transport facilities and services, i.e., sustainable planning, investment and development.

The proposed model is significant for the reason that it is the result of research that is rare in Europe and unique in the SEE region. For this reason, the research results could not be compared with other models formed for small and medium-sized cities in the region and Europe.

The proposed model was developed and tested for small and medium-sized cities in the SEE region. Testing of models for large and mega cities was not carried out for the reason that no basic research was carried out on them. The model was formed based on research conducted on a weekday in normal traffic conditions, including sunny weather, without heavy precipitation, wind, etc., and in accordance with the established practice in this area. [5].

Besides the introduction section, the paper has been structured in the following five parts. The second section gives emphasis on the basic objectives, motivation, and purpose of the performed study, while the third section represents background in areas of exploration. The fourth part of the paper shows and explains the research methodology with a clear diagram of the research flow, performances of shopping centres, details related to data collection, performing experiments, etc. In the fifth section, results are shown in detail, and a discussion has been provided. The last section summarizes the main aspects of the paper with guidelines for future research.

2. Basic Objectives and Purpose of the Work

The main goal of this paper is to determine and define the parameters and indicators in real traffic conditions based on the results of the research, according to which the model of traffic demand generated by shopping centres of certain characteristics will be defined in the analysis procedure. The model is sensitive to local impacts because it includes parameters and indicators determined by research in real, local traffic conditions, which is also one of the goals of the work. Based on the proposed model, traffic demand generated by the shopping centre can be determined, depending on the characteristics of the location, the characteristics of the shopping centre (surface of the sales area and the number of parking spaces), the number of inhabitants of the town, the number of households and the number of registered passenger vehicles. Model testing was performed for a randomly selected shopping centre, and data on the size of traffic demand was collected by research in local conditions. Namely, the application of the model in the procedures for assessing the impact on traffic in real conditions and the environment is also one of the goals of the research.

The output results of the application of the model can be used to analyze the impact of the construction of shopping centres at selected locations on the traffic conditions on the surrounding street and road network in medium-sized towns of the Southeast Europe region, which is the main purpose of the paper. Therefore, the proposed model for determining traffic demand can serve to select the optimal location for the construction of the shopping centre, i.e., the location that will generate the least harmful negative impacts on traffic conditions. In the further procedure, and in accordance with the size of the generated traffic and its impact on the traffic conditions, the results obtained by the model can be used to dimension the elements of the street network, i.e., intersections and segments of roads, that is, links between intersections.

One of the objectives of the paper is to develop models for an objective assessment of the generated traffic demand for significant land-use changes, such as the construction of shopping centres in mid-sized towns, which would be a step in the right direction to promote reliable traffic planning and promote the adoption of TIAs within the construction permitting process [6]. In Bosnia and Herzegovina and the Republic of Serbia, such a process has not yet been systematically established. Namely, the systematic assessment of the generation of travel for land use changes is not a mandatory procedure in Bosnia and Herzegovina, or Serbia, so it has not been applied so far in the procedures for selecting the optimal location of the shopping centre, with the exception of several individual cases, at the request of private investors.

In addition to the application in the planning analysis procedures, the model can also have an application in the operational analysis process. Namely, the model can be applied to determine the newly created traffic, i.e., the increase in the existing traffic demand in the event of a change in the purpose of the land in the previously built parts of urban areas, i.e., the construction of the planned shopping centre. It is often difficult or impossible to change the profile of roads, i.e., the constructed street network, so the results of the application of the model are necessary in order to analyze the effects of the application of various measures in order to preserve the existing level of service. Therefore, the implementation of the model will indirectly enable the possible negative consequences on the traffic conditions of the street network due to the construction of the shopping centre to be prevented at the planning stage. By applying the model, based on the assessment of the increase in traffic demand, the effects of pollutant emissions can be analyzed, which can serve as an additional argument for decision-makers.

In order to achieve the set goals, the research was conducted in several locations where there are built and functioning shopping centres, of different sizes and in different medium-sized towns in the Republic of Serbia and Bosnia and Herzegovina and with different population numbers, in order to explore the connection between the characteristics of shopping centre facilities, as well as certain socioeconomic and demographic characteristics of the area in which they are located, with the volume of traffic generated by such facilities. The decision to conduct the analysis and research in medium-sized towns was made due to the fact that in the Republic of Serbia and Bosnia and Herzegovina, there are not many towns whose population exceeds 200,000. Namely, there is only one city in Bosnia and Herzegovina, and in the Republic of Serbia there are three cities with over 200,000 inhabitants, and the situation is similar in other countries in the region where the developed model is expected to be applied. The survey was not conducted in smaller towns, i.e., towns with up to 50,000 inhabitants, since it can be assumed that the traffic demand generated by shopping centres is not a problem and does not have a significant impact on the traffic conditions in the surrounding road and street network. The locations covered by the survey are diverse in terms of their characteristics, primarily according to their position in the town or city (distance from the centre), the size of the facilities themselves (GLA—gross leasable area), geometry and number of accesses to the facilities, the number of available parking spaces and the facilities in the facilities themselves.

3. Literature Review

The broadest internationally recognized reference and best-known source associated with travel generation equations and rates worldwide is the Travel Generation Handbook published and regularly updated by the Institute of Transportation Engineers—ITE. Travel generation studies have been conducted since the 1960s in the US and Canada. The first edition of the ITE manual in 1976 made recommendations on the rates of generating travel for 50 land uses, which are the result of a survey of 500 traffic studies. In 2021, the ITE published the 11th edition of the manual, which contains data on travel generation rates for over 170 land uses determined based on the results of 5000 studies. For the purposes of estimating trip generation, an independent variable is defined as a physical, measurable, and predictable unit that describes a research site. The most commonly used parameters in the ITE database are gross floor area available to visitors (GFA), gross leasable area (GLA), number of seats, number of employees and number of housing units. In addition to this handbook, in the US some individual countries and areas have established local trip generation rates based on perceived and systematized characteristics in their local communities, which are used when approving land development and land use change. So, for example, in the state of Texas, there is a Texas trip generation manual (2014) similar to the ITE manual but based on local Texas data. The results of the study [7] revealed that Texas trip generation rates were generally lower than comparable rates in The ITE handbook. Also, The ITE Travel Generation Manual has been criticized for overestimating vehicle traffic in its application in urban areas [8,9,10]. In connection with the same, research was carried out to adjust ITE estimates of vehicle traffic for urban units [11].

It is crucial for developing countries to have their own manual for local data [12], as using international data for local conditions can lead to errors in determining travel rates, as well as estimating the origin of travel [13]. Errors in travel rates and generation of travel result in the application of inadequate mitigation measures [14].

TRICS (Trip Rate Information Computer System) [15] is a national standard computerised travel generation analysis system for the UK and Ireland and is an integral and essential part of the traffic assessment process. First launched in 1989 by six counties of the South East of England, it has expanded through continuous investment and development into a comprehensive database to analyze the emergence of travel and is now maintained and managed by a team of TRICS Consortium Ltd. management and technical staff, based in Barnet, London. The system allows its users to establish potential levels of trip generation for their development scenarios using a series of database filtering processes. The TRICS database includes over 8000 traffic surveys covering 121 separate land use classifications. In addition to inbound and outbound traffic and multi-modal counting (covering a wide range of separate counting types and modes), TRICS site records include comprehensive descriptive details about the local and site environment, information about the size, composition, and functions of the site, and details about on-site and off-site parking. Most land uses have one to four variables, or parameters, that can be used to calculate travel rates. GFA, headcount, parking spaces, and site area apply extensively to a wide range of land uses when calculating travel or parking rates. The most common parameter fields in the TRICS database are GFA, parking spaces and site area [16].

The most important source relating to travel generation in Australia is The New South Wales Roads and Traffic Authority’s 2002 Guide to the Development of Traffic Generation (RTA). The RTA database contains information on the travel rates of vehicles for 34 land uses divided into nine categories. The database was created in 1984 and was last updated in 2002. The most commonly used parameters for The RTA database are gross floor area available to visitors (GFA) and housing units [17].

In New Zealand, the New Zealand Trips Database Bureau—TDB—provides a travel-generating database containing 693 New Zealand locations and 192 Australian RTA locations. Travel rates can be calculated using various parameters or data fields. The most common rate is per 100 m² gross area—GFA [16].

The two established travel-generating manuals most commonly used in South Africa are the Committee of Transport Officers Handbook [18] and the National Ministry of Transport Handbook [19]. According to Mogakabe et al. [20], the travel rates used in the COTO manual are based on insufficient data, and in some cases international data that may not necessarily be applicable to South African traffic conditions. Therefore, there is a need to develop travel generation rates that are based on local data and comparable to international data in South Africa. The NdoT was originally adopted in 1989 and provides an estimate of the number of expected trips for a range of land uses. It was reproduced in 1995 and is now widely used. This document identifies the trip production rates for the various movements and factors affecting the trip production. In the doctoral dissertation [21], a comparison was made between the actual rates of generating trips for shopping centres in low-income settlements and the accepted rates of generating trips from the manual (NdoT and Coto). The conclusion is that the difference between the actual recorded travel rate for the three shopping centres in the low-income area and the Coto rate is less than 10%, which is considered an acceptable degree of accuracy when generating travel for new development is envisaged, while the NDoT rates are not considered appropriate for these developments, as they are significantly higher (up to 77%) than the actual recorded rates. The conclusion is that the trip generation standards used are not always compatible with the South African context and that the existing trip generation standards need to be updated.

The Abu Dhabi Department of Transportation (DoT) has developed the Abu Dhabi Road and Transit Authority [22] for several land uses in the United Arab Emirates-based on best international practices and extensive research and surveys conducted through development trends in the Emirates (analyses and research from nearly 400 different locations across the Emirates). Trip generation rates are determined for each land use for three periods during the day (morning, noon and evening) to make the manual compatible with other international handbooks, and in general, the methodology is similar to the ITE methodology.

In Brazil, a certain number of studies have been conducted to estimate the volume of travel generated by shopping centres. Articles [23,24] were developed based on data collected in shopping centres in the city of Sao Paolo, while Grando [25] and Goldner [26], in their articles, used the data obtained from the shopping centres mostly in big towns in different regions of the country. Apart from the above-mentioned studies, certain research studies for the assessment of the volume of travel generated by shopping centres in Brazil were conducted by other authors [27,28,29]. In all mentioned articles, when estimating the number of vehicles attracted by shopping centres, only the size of the building, i.e., the gross leasable area or the total built-up area, is taken into account

In different individual surveys around the world, many travel generation studies have been carried out, focusing on specific types of land use. So, among other things, there are studies that have focused on shopping centres. The 1977 New Hampshire Department study was one of the first to focus on trips generated by Meena and Patil [30]. The study proposed two different models. The first model refers to a large shopping centre (GLA greater than 200,000 square feet) in which only GLA is found as an independent variable, and the second model is presented for a medium-sized shopping centre (usually for 50,000 to 150,000 square feet) in which GLA, the age of shopping centres and traffic on surrounding roads appear as independent variables that affect the volume of traffic generated by shopping centres. Through the work [31], an analysis of a travel generation study conducted for ITE and the International Council of Shopping Centres (ICSC) in 1994 and 1995 was presented. The study explored several different aspects of trip generation in shopping centres, including the relationship between trip generation and combinations of several independent variables, concluding that gross leasable area (GLA) is significant for predicting trip generation for a shopping centre. Also, in the paper [32], it was concluded that the gross leasable area (GLA) is significant for predicting the emergence of travel for shopping centres.

Al Masaeid et al. [33] developed a travel and parking generation model for shopping centres and found that gross area and the number of employees are the most significant independent variables. Faghri et al. [34] developed an alternative methodology based on artificial neural networks (artificial neural networks—ANN) to analyze the relationship between the percentage of transient trips generated by shopping centres and the factors affecting them and compare it with the regression model results from the American Institute of Transportation Engineers’ (ITE) Travel Generation Handbook. Through their work, it was found that ANN-based models have the ability to more accurately represent the relationship between the percentage of transient trips and independent variables than traditional regression models. Uddin et al. [10] conducted tests in six shopping centres in the Dhanmondi area of Dhaka (Bangladesh) and concluded that the physical characteristics of the shopping centre (total number of parking spaces, gross area and number of stores) are significant for predicting the occurrence of travel.

Mamun et al. [2] collected data from six medium and small shopping centres in Dhaka and developed two macroscopic models to generate travel. The first model has a gross available area (GFA), the number of parking spaces and the number of restaurants as significant independent variables in determining the attractiveness of trips to the mall, while the second model uses the total number of stores instead of the GFA. It is concluded that the total number of stores is a better predictor than the GFA for determining the attractiveness of trips to the shopping centre.

Kikuchi et al. [35] conducted a study in 18 shopping centres in Delaware (USA) to determine travel attractiveness rates in shopping centres. This study adopted two approaches, microscopic and macroscopic, to calculate the rate of travel attraction. The microscopic approach deals with the relationship between the travel attractiveness rates of individual stores and the shopping centre as a whole, while the macroscopic approach connects the attractiveness of shopping centre travel as a function of the physical characteristics of the shopping centre itself (total number of parking spaces, total area of establishments and number of establishments). The conclusion is that the microscopic model gives more precise results compared to the macroscopic approach.

In their study, Meena and Patil [30] developed models for generating trips for shopping centres in Mumbai Metropolitan (India). In total, eighteen shopping centres were selected to develop a travel generation model using regression analysis. It was found that the generation of trips is related to five parameters: built-up area, number of screens in the multiplex, number of seats in the multiplex, number of kiosks and number of stores in shopping centres. It was found that other parameters such as gross leasable area, number of parking spaces, population density and employment density do not have a significant impact or correlation with the number of trips generated from shopping malls. Also, from this study, it can be concluded that the built-up area (BUA) is a better parameter than the gross leasable area (GLA) to model travel generation in the context of shopping centres.

George and Kattor [36] developed an appropriate travel attractiveness model in their study using multiple regression analysis, to predict future trips attracted by a commercial node of certain characteristics for a mid-sized city in Kerala (India). Similar to related studies, this study also found that the number of employees is highly correlated with the attractiveness of travel, and the total commercial area is moderately correlated with the attractiveness of travel. Other factors found to be related to travel attractiveness are the number of commercial establishments, the percentage of stores selling food items, the percentage of offices, shops and banks in a commercial hub, and the percentage of commercial establishments with only one floor and more than two floors.

In the paper [37], travel generation rates were determined for commercial centres, four shopping centres in Lima Norte (Peru), and then the resulting travel generation rates were compared with indices from ITE and with case rates in Venezuela. The independent variables used are the total area, the leasing area, the number of parking lots and the number of individual business premises, and the dependent variables are the number of private vehicles and the number of pedestrians visiting shopping centres. For vehicle mode, the variable most appropriate is the number of parking spaces. When comparing the models of rates developed in this paper with the rates of other countries (ITE and Venezuela), it is concluded that the rates are significantly below ITE by about 60% less and by approximately 40% more than in Venezuela.

In their study, Majeed and Qasim [38] established mathematical models for predicting travel attraction for the sector Bab al-moadam in the city of Bagdad which include the characteristics of land use. The study included educational centres, state institutions, shopping centres and healthcare institutions, that is, models of travel attraction were developed for different land uses. The conclusion is that the travel attraction forecasting model regarding shopping centres significantly depends on the number of shops in relation to other factors included in the analysis since it was established that the number of shops significantly affects the attraction to visit shopping centres.

In the research of Palani and Malarvizhi [39] the factors of traffic attraction in different classes of shopping centres in India were identified, measured by consumer psychology. The survey was conducted in 11 different shopping centres. In the above-mentioned study, it was concluded that more factors related to socio-demographic and travel characteristics of shopping centres users can affect the traffic attractiveness of the shopping centre. Furthermore, it was concluded that all factors did not affect all the attractions of the shopping centre. The factors differed depending on the classes of the shopping centres, which were classified into four clusters according to the gross leasable area and the facilities which seem to be dominant characteristics of the shopping centre attractiveness. Das and Ray [40] conducted surveys in six shopping centres in the town of Chattogram (Bangladesh) and concluded that the physical characteristics of a shopping centre (the total number of parking lots, area, number of shops and the number of employees in the shopping centre) are significant for travel forecasting.

4. Research Methodology

This paper defines a research method based on known rules and methods of scientific knowledge, i.e., on methods of data collection and database formation, methods of analysis, synthesis, induction, generalization, classification, description and statistical and mathematical methods. Figure 1 shows the diagram which describes the complete research flow by phases.

As part of the conducted research, a database was formed, containing collected data from several locations with previously built shopping centres that are in operation, which generate additional flow requirements on the surrounding street network on a daily basis. The survey was conducted in several medium-sized towns in two countries of the Southeast Europe region (the Republic of Serbia and Bosnia and Herzegovina). The research included the collection of data on the basic characteristics of the observed shopping centres, as well as the characteristics of the environment in which they are located. Given the complexity of the research, which required the engagement of significant human and financial resources, in addition to their own research, available statistical data related to the demographic and socioeconomic characteristics of the towns in which the research was conducted were used.

The shopping centres that were the subject of research are not equipped with modern ITS systems for the detection of vehicles, pedestrians, and cyclists, so the traffic demand was determined by counting and recording in pre-prepared sheets. All surveys were conducted on a working day. Modern ITS technologies enable the formation of traffic demand databases for any period.

The basic parameters used in this paper, required to determine the volume of traffic that will be generated by the construction of shopping centre facilities in a town, and thus determine its impact on the traffic conditions of the surrounding road and street network, are the number of inhabitants of the town, the number of households, the degree of motorization, the position of the facility in the town (whether they are located in the urban part of the town or suburbs-distance from the centre), the size of the facilities (total surface sales area), the number of stores, their content and the number of available parking spaces. All these parameters were obtained by local measurements at selected locations and by collecting known statistics. For the purposes of the research, medium-sized towns (from 50–200 thousand inhabitants) were selected from the entire territory of the Republic of Serbia and Bosnia and Herzegovina with different numbers of inhabitants, in different locations of built shopping centres in terms of their position (in the urban part of the town or city, or in the suburbs).

In addition to the data that were publicly available, the collection of data on the analyzed parameters was performed by the method of counting and on-site measurements at all selected locations within the survey area. The counting was carried out with the aim of determining the specific characteristics of each of the selected locations in terms of the traffic generation rate through the number of arrivals and departures to the observed locations by passenger vehicles at 15 min intervals during the day, the number of users of the shopping centre services in accordance with the number of passengers in the vehicles and the degree of occupancy of the parking space.

For each location, the available number of parking spaces and the size of the facility were determined, i.e., the total area of the sales area and the number of stores. Counts and measurements at selected locations were performed for two days, in May 2021, on May 20 and 27 (Thursday), throughout the day, i.e., during the opening hours of shopping centres. In accordance with opening hours, for one part of the shopping centres, the research was carried out in the period from 09:00 to 21:00, and for the other part from 10:00 to 22:00. By processing the collected data, the parameters necessary for further analysis were obtained, which are the flow of vehicles at 15 min intervals, at the entrance and exit from the parking lot at the observed locations of shopping centres, the number of passengers in vehicles and the occupancy of parking spaces.

By analyzing the results of the research, the number of passenger vehicles that appeared at the entrance to the parking area of shopping centres at the time of the research was obtained, which served to determine the volume of generated traffic. All surveys carried out for the purpose of drafting this article are related to the real state of the size of traffic generated by shopping centres, so the results obtained in the investigated sample are sufficient for the obtained models to be applied in all conditions of local traffic and at the locations of all types of shopping centre facilities in medium-sized towns.

In order to collect relevant data that would serve to form an appropriate database, a survey was conducted at 8 locations, i.e., in zones where there are built and functioning shopping centres in eight towns (Banja Luka, Tuzla, Bijeljina, Kragujevac, Šabac, Sombor, Sremska Mitrovica, and Lazarevac) in two countries, Bosnia and Herzegovina and the Republic of Serbia (Figure 2).

The locations covered by the survey are diverse in terms of their characteristics, primarily according to their position in the town (distance from the centre), the size of the facilities themselves (GLA—gross leasable area), geometry and number of accesses to the facilities, the number of available parking spaces and the facilities in the facilities themselves.

Table 1. Names and locations of the shopping centres where the research was conducted.

Location No.	Name of the Facility:	Town	Address	Location	Government (Entity/Area)
loc. 1	Delta Planet—Shopping Mall	Banja Luka	Bulevar srpske vojske 8	44°46′47.96″ N 17°12′24.99″ E	Bosnia and Herzegovina Republic of Srpska
loc. 2	Bingo City Center	TUZLA	Mitra Trifunovića Uče 2.	44°31′56.58″ N 18°39′7.27″ E	Federation of Bosnia and Herzegovina
loc. 3	City Mall	Bijeljina	Sremska 10	44°45′46.05″ N 19°12′22.32″ E	Bosnia and Herzegovina Republic of Srpska
loc. 4	Plaza Shopping Centre	Kragujevac	Bulevar Kraljice Marije 56	44° 0′30.98″ N 20°53′46.54″ E	Republic of Serbia ŠUMADIJA
loc. 5	Capitol Park	Šabac	Zapadna Transverzala I, No.4	44°45′14.51″ N 19°39′46.68″ E	Republic of Serbia (Mačva)
loc. 6	Capitol Park	Sombor	Strosmajerova 28	45°45′53.00” N 19° 7′29.19″ E	Republic of Serbia (Western Bačka)
loc. 7	STOP SHOP	Sremska Mitrovica	Bul. Konstantin Veliki 74	44°58′35.72″ N 19°35′29.56″ E	Republic of Serbia (Srem)
loc. 8	STOP SHOP	Lazarevac	Železnička 4	44°23′13.07″ N 20°15′5.32″ E	Republic of Serbia Belgrade

provides a list of the names of shopping centres and their exact locations, while

Table 2. Characteristics of the shopping centres where the research was conducted.

Location No.	Name of the Facility:	Total Sales Area-GLA (m2)	Parking Spaces	Distance from the City Centre (km)	Working Hours
loc. 1	Delta Planet—Shopping Mall	30,200	970	2.4	9–21
loc. 2	Bingo City Center	20,500	415	2.4	9–21
loc. 3	City Mall	11,350	270	1.5	9–21
loc. 4	Plaza Shopping Centre	22,000	610	1.8	10–22
loc. 5	Capitol Park	8800	445	3.5	10–22
loc. 6	Capitol Park	4000	248	1.5	9–21
loc. 7	STOP SHOP	8400	421	2.4	10–22
loc. 8	STOP SHOP	10,500	215	1.3	10–22

provides an overview of the characteristics of all shopping centres that were used as parameters for analysis.

Defining the area of influence as a geographical term in terms of defining the area from which shopping centres attract the majority of customers is extremely important because it is one of the basic steps in assessing the impact on the transport system. According to Mussi et al. [41], the area of influence is defined as the geographical area where shopping centres will attract the population to make purchases or meet their needs for entertainment and services. According to ULI [42], the term “shopping area” or “market area” is usually defined as the area from which the largest share of the continuous clientele necessary for the permanent maintenance of the shopping centre is obtained. In this way, the number of inhabitants is obtained as one of the changeable variables in the assessment of the generation of travel conditioned by the construction of a shopping centre, and thus for the assessment of the impact on traffic (TIA) by such construction.

Different authors differently set criteria for delineating and defining areas of influence, thus the areas from which additional traffic is generated, caused by construction and the existence of a specific shopping centre. There are a lot of different definitions of areas of influence adopted by different authors.

Thus, Grando [25] adopts [43] as the criterion by which the area of influence of a shopping centre is defined as a geographic area from which most customers are attracted, and 93.3% of travel is additionally divided into primary, secondary and tertiary areas, while the remaining 6.7% of travel is outside the area of influence. Likewise, some variables are added, which should be considered when defining the area of influence, such as travel distance, travel time, physical obstacles, distance from the town centre and major competition and the attractiveness of the future shopping centre.

The area of influence of the shopping centre [44] is defined as the area which represents the physical demarcation of the range so as to meet most of their needs and from which most customers are attracted, where 83% of travel is additionally divided into a primary, secondary and tertiary area, while the remaining 17% of travel outside the area of attractiveness. The shopping centre customer from the primary area reaches the shopping centre within 10 min by car, from the secondary area within 10 to 20 min by car, while from the tertiary area, the customer reaches the shopping centre within 20 to 30 min by car.

Furthermore, according to Goldner [26], the area of influence emphasizes what represents the geographical area from which most customers originate, thus their travel to shopping centres is limited by equal timelines every 5 min (from 20 to 30 min) and the lines are of equal distance, drawn every 1 km in the form of concentric circles whose centre is the shopping centre (usually up to 8 km). Here, the position of the facility itself differs, whether they are within the urban part of the town or outside the town area. Therefore, for the shopping centres positioned within the urban part of the town, the area of influence represents the area from which 98.8% of travel is conducted, according to the shopping centre additionally divided into primary, secondary and tertiary area, and the remaining 1.2% of travel is outside the area of influence. For those shopping centres which are positioned outside the urban area, the area of influence represents the areas from which 86.7% of travel is directed to the shopping centre and the remaining 13.3% of travel is outside the area of influence.

ULI [45] defines the primary area as travel time by car of up to 10 min with a distance of up to 10 km, the secondary area as travel by car of up to 20 km for 20 min, and the tertiary area with a travel time of up to 30 min and a distance up to 30 km.

Conceição [46] defines the area of influence as the area in which there are most customers who will purchase in the shopping centre. With such a definition, he takes into consideration travel duration, natural obstacles, the distance and the potential (activity) of the future shopping centre. In accordance with the criteria adopted by [45], he also divides this area into three zones (areas) of influence: primary, secondary and tertiary areas.

Keefer [43] does not differ between primary, secondary and tertiary areas, but the area of influence is defined as the distance up to 8 km from the shopping centre if traveling by car for up to 20 min.

Martins [27] uses more criteria for defining the area of influence, taking into consideration the position and the characteristics of the shopping centre. He uses one of the criteria to define the area of influence as the area from which 95% of travel is conducted towards the shopping centre, and the remaining 5% is outside the area of influence. This is related to the shopping centres in commercial zones, that is, outside residential zones.

According to Correa and Goldner [47], the area of influence is divided into primary, secondary and tertiary. In the primary area, the user of the shopping centre reaches it within 5 to 10 min by car, depending on traffic conditions and access roads to the area. In the secondary area, the user of the shopping centre reaches it within 10 to 20 min by car depending on traffic conditions and access roads, without competitive facilities. In the tertiary area, the user of the shopping centre reaches there within 20 to 30 min by car depending on traffic conditions and access roads, with the existence of highly competitive local shops like hypermarkets and supermarkets, and other shops, in general.

The given criteria for delineating and defining the area of influence imply certain deviations when taking into account and comparing variables related to distance from the shopping centre time of travel by car, travel distribution and the location of the shopping centre itself. Analyzing the various above criteria, it can be concluded that the maximum coverage of the impact area is up to 30 km from the shopping centre, up to 30 min of travel by car and from which 98% of trips directed to a specific shopping centre arise.

By delineating and defining the area of influence for the shopping centres covered by this research, it can be concluded that for all shopping centres except for the Bingo City Center Tuzla shopping centre, the entire territory of the town in which they are located can be considered as the area of influence.

Namely, observing the positions of shopping centres in each of the towns covered by the research, except Tuzla, it is concluded that apart from the territory covered by the administrative borders of the cities themselves, there is no significant settlement outside that territory that would be located according to the presented criteria for limiting the area of influence, up to 30 km from the shopping centre and up to 30 min by car.

For the town of Tuzla, i.e., the Bingo City Center, the situation is different from all other cities and shopping centres covered by the research. The shopping centre itself is located in the western part of the town next to the town motorway, which continues and divides into two main directions. The first continues west to the municipality of Lukavac, which borders Tuzla and whose administrative centre is 13 km away from the given shopping centre, with a car ride of 17 min. The second main route from the shopping centre continues south to the municipality of Živinice, which also borders Tuzla and whose administrative centre is 13 km away from the given shopping centre, with a car ride of 18 min. As both municipalities are significant settlements and they are located within the area of influence defined by the previously described criteria, for the Bingo City Center shopping centre, the area of influence through this research is defined as the complete administrative territories of the town of Tuzla and the municipalities of Lukavac and Živinice.

Officially available data were used to determine the number of inhabitants (households) in the towns covered by this survey. As the last census in Bosnia and Herzegovina (Federation of Bosnia and Herzegovina and Republic of Srpska) was conducted in 2013, and in the Republic of Serbia in 2011, for the purposes of this survey, the population estimate for 2020 was used based on the results of the last censuses and on the results of statistical processing of natural and mechanical population movements [48,49,50].

In addition to the number of inhabitants (households) as changeable variables in estimating the generation of travel conditioned by the construction of a shopping centre, the number of registered passenger vehicles (cars) was taken as another variable. The number of registered passenger vehicles in the towns of Bosnia and Herzegovina used through the analysis in this paper is by 1 May 2021 [51], while the available number of passenger vehicles in the towns of the Republic of Serbia used through the analysis in this paper is by 1 January 2021 [50].

An overview of the towns in which the surveys were conducted, by country and with their characteristics related to the number of inhabitants, number of households and number of registered passenger vehicles is presented in

Table 3. Overview of the surveyed towns with their general characteristics.

Town	Government	Population	Number of Households	Number of Registered Passenger Vehicles
Banja Luka	Bosnia and Herzegovina	185,094	65,010	76,769
TUZLA (Tuzla, Lukavac, Živinice)	Bosnia and Herzegovina	210,307 (109,527 + 42,927 + 57,853)	77,717 (42,630 + 16,656 + 18,431)	68,242 (36,749 +13,560 +17,933)
Bijeljina	Bosnia and Herzegovina	103,783	34,309	39,319
Kragujevac	Republic of Serbia	175,716	59,991	56,157
Šabac	Republic of Serbia	109,340	39,091	35,719
Sombor	Republic of Serbia	77,463	31,730	22,451
Sremska Mitrovica	Republic of Serbia	74,609	27,218	22,838
Lazarevac	Republic of Serbia	56,125	18,862	20,829

.

5. Research Results and Discussion

This paper analyzes the relationship between the size of traffic attracted by the zone of attraction—the shopping centre and the characteristics of the location itself (the surface of the sales area and the number of parking spaces), the number of inhabitants of the town, the number of households and the number of registered passenger vehicles. The aim of the analysis is to form a mathematical model (relation) on the basis of which it is possible to calculate the increase in traffic in relation to the characteristics of the attraction zone as an element that affects the conditions of traffic on the road and street network.

Within the subject analysis, descriptive statistics of parameters relevant to the formation of models were presented, the impact of certain parameters on traffic attracted by the attraction zone was examined, and a mathematical relation was derived on the basis of which it is possible to predict the total daily volume of traffic generated in the zones of shopping centres, as well as the total number of people who visit the shopping centre during the day with vehicles (cars).

5.1. Model for Estimating the Total Daily Number of Visitors and Vehicles in the Shopping Centre Zone

The formation of a model to estimate the total number of visitors, i.e., the total number of vehicles attracted by the attraction of the shopping centre zone, is based on a linear regression analysis. Models based on linear regression start from the following hypotheses:

Hypothesis 1.

(Null Hypothesis) There is no linear dependence with the given dependent variable between the independent variable values figuring in the set regression equation. The null hypothesis is accepted if the probability for the accuracy of the null hypothesis (p-value) is higher than the given significance threshold (α), otherwise the null hypothesis is rejected.

Hypothesis 2.

(Alternative Hypothesis) An assertion contrary to that defined in the null hypothesis. An alternative hypothesis is not rejected when the probability for the accuracy of the null hypothesis (p-value) is less than the given significance threshold (α).

Normal values of the significance threshold α range in the interval 0.15 ≥ α ≥ 0.05. For the formation of the given model, the adopted value of the significance threshold is α = 0.05, i.e., a 95% confidence interval is accepted.

Initial model limitations:

• the model refers to visits to the shopping centre by a passenger car,
• the model can be applied to towns between 56,125 and 210,307 inhabitants,
• the model can be applied to towns between 18,862 and 77,717 households in size,
• the model can be applied to towns with a total number between 20,829 and 76,769 registered passenger vehicles,
• the model can be applied in cases where the total sales area of the shopping centre has a value between 4000 and 30,200 m²,
• the model can be applied in cases where the total number of parking spaces of the shopping centre has a value between 215 and 970.

Linear regression analysis was performed in IBM SPSS Statistics 22 software environment. Two regression models were formed:

(1)

Model for estimating the total daily number of visitors who visit the shopping centre using vehicles,

(2)

Model for estimating the total daily number of vehicles in the shopping centre zone accessing the parking facility.

Out of a total of eight observed locations, the characteristics determined on seven facilities were used as the basis for the formation of the model, one randomly selected location was used to test the model.

Selection of independent variables was performed by the Backward Elimination algorithm, which starts by including all variables in the equation/model, and then their successive removal is performed. The variable with the lowest partial correlation with the dependent variable is considered as the first to be eliminated if it meets the elimination criterion, which by default is set to a probability F value equal to 0.1. If the F probability obtained is higher than the set criterion, the variable is removed. The next variable in the equation with the least partial correlation with the dependent variable is considered for removal. The procedure is stopped when there are no variables in the equation that meet the removal criteria.

In order to simplify the work, the five observed independent variables are marked in the regression analysis as follows:

• Population—BS
• Number of households—BD
• Number of registered passenger vehicles—BRPV;
• Total sales area—UPP;
• Parking spaces—BPM

The two dependent variables have the following labels:

• Total daily number of persons in vehicles on arrival (drivers and passengers together) visiting the shopping centre using vehicles—UBO;
• The total daily number of vehicles in the shopping centre zone accessing the parking lot of the facility—UBV.

The previously described Backward Elimination procedure, to estimate the total daily number of visitors in the shopping centre zone (UBO), formed five models whose summary results are shown in

Table 4. Linear models for UBO estimation—summary results.

Model	Independent Variables/Predictors	R	Square	Adjusted R Square	Std. Error of the Estimate
1.	BPM, BD, UPP, BRPV, BS	0.986	0.973	0.838	664.377
2.	BD, UPP, BRPV, BS	0.983	0.966	0.897	529.279
3	BD, UPP, BS	0.972	0.946	0.891	544.736
4	BD, UPP	0.934	0.872	0.808	722.491
5	UPP	0.895	0.801	0.761	806.336

. For each model, the correlation value—R, the coefficient of determination—R Square, the adjusted coefficient of determination—Adjusted R Square and the standard error—Std. Error of the Estimate.

Taking into account the values of the above-mentioned model quality indicators, it is concluded that Model 1 with all five independent variables shows the best agreement with the actual data (R Square = 0.973).

The results of the analysis of variance (ANOVA) for the five UBO assessment models are shown in

Table 5. ANOVA for UBO linear regression models.

Model		Sum of Squares	DF	Square	F	Sig.
1.	Regression	15,894,707.153	5	3,178,941.431	7.202	0.275
	Residual	441,396.561	1	441,396.561
	Total	16,336,103.714	6
2.	Regression	15,775,831.402	4	3,943,957.850	14.079	0.067
	Residual	560,272.313	2	280,136.156
	Total	16,336,103.714	6
3	Regression	15,445,891.989	3	5,148,630.663	17.351	0.021
	Residual	890,211.725	3	296,737.242
	Total	16,336,103.714	6
4	Regression	14,248,130.187	2	7,124,065.094	13.648	0.016
	Residual	2,087,973.527	4	521,993.382
	Total	16,336,103.714	6
5	Regression	13,085,214.837	1	13,085,214.837	20.126	0.006
	Residual	3,250,888.877	5	650,177.775
	Total	16,336,103.714	6

. The realised significance of the Sig. test statistics indicates that only linear models 3, 4 and 5 have statistically significant differences in independent effects on the dependent variable.

The specific values of the regression coefficients figuring in the models—Unstandardized Coefficients B are given in

Table 6. Linear model coefficients for UBO estimation.

Model		Unstandardized Coefficients		Standardized Coefficients	T	Sig.
		B	Std. Error	Beta
1.	Constant.	791.404	899.357		0.880	0.541
	BS	−0.138	0.072	−5.206	−1.920	0.306
	BD	0.457	0.226	6.097	2.207	0.292
	BRPV	−0.128	0.127	−1.819	−1.004	0.499
	UPP	0.337	0.171	1.925	1.970	0.299
	BPM	1.368	2.636	0.211	0.519	0.695
2.	Constant.	1109.700	524.016		2.118	0.168
	BS	−0.130	0.056	−4.891	−2.323	0.146
	BD	0.410	0.164	5.466	2.494	0.130
	BRPV	−0.088	0.081	−1.261	−1.085	0.391
	UPP	0.319	0.134	1.824	2.391	0.139
3	Constant.	1300.942	507.910		2.561	0.083
	BS	−0.107	0.053	−4.039	−2.009	138
	BD	0.301	0.134	4.015	2.247	0.110
	UPP	0.191	0.063	1.090	3.012	0.057
4	Constant.	1367.230	672.225		2.034	0.112
	BD	0.034	0.023	0.454	1.493	210
	UPP	0.092	0.053	0.528	1.734	0.158
5	Constant.	1963.657	603.301		3.255	0.023
	UPP	0.157	0.035	0.895	4.486	0.006

, and the statistical significance of each of them is represented by the value of the Sig. t test. It can be observed that the constant (Sig. = 0.023) and the coefficient with the UPP Model 5 variable (Sig. = 0.006) are the only statistically significant coefficients.

The formed models for the estimation of the UBO (total number of persons in the vehicle on arrival—both driver and passengers) can be mathematically recorded in the manner presented in

Table 7. UBO estimation models—total number of occupants on arrival.

MODEL	Mathematical Equation of the Model
Model 1	$UBO=791.404-0.138·\mathrm{B}\mathrm{S}+0.457·\mathrm{B}\mathrm{D}-0.128·\mathrm{B}\mathrm{R}\mathrm{P}\mathrm{V}+0.337·\mathrm{U}\mathrm{P}\mathrm{P}+1.368·\mathrm{B}\mathrm{P}\mathrm{M}$
Model 2	$UBO=1109.7-0.130·\mathrm{B}\mathrm{S}+0.410·\mathrm{B}\mathrm{D}-0.088·\mathrm{B}\mathrm{R}\mathrm{P}\mathrm{V}+0.319·\mathrm{U}\mathrm{P}\mathrm{P}$
Model 3	$UBO=1300.942-0.107·\mathrm{B}\mathrm{S}+0.301·\mathrm{B}\mathrm{D}+0.191·\mathrm{U}\mathrm{P}\mathrm{P}$
Model 4	$UBO=1367.230+0.034·\mathrm{B}\mathrm{D}+0.092·\mathrm{U}\mathrm{P}\mathrm{P}$
Model 5	$UBO=1963.657+0.157·\mathrm{U}\mathrm{P}\mathrm{P}$

.

In the same way, five linear models were created to estimate the total number of vehicles that entered—accessed the parking facility (UBV), the summary results of which are shown in

Table 8. Linear models for UBV estimation—summary results.

Model	Independent Variables/Predictors	R	Square	Adjusted R Square	Std. Error of the Estimate
1.	BPM, BD, UPP, BRPV, BS	0.984	0.969	0.812	447.353
2.	BD, UPP, BRPV, BS	0.983	0.965	0.896	332.262
3	BD, UPP, BS	0.962	0.926	0.852	397.417
4	BD, UPP	0.918	0.842	0.764	501.617
5	UPP	0.883	0.780	0.736	529.921

. The Model 1 has the highest values of R, R Square and Adjusted R Square, in which all five independent variables are represented. Based on the results shown in

Table 9. ANOVA for UBV linear regression models.

Model		Sum of Squares	DF	Square	F	Sig.
1.	Regression	6,187,334.644	5	1,237,466.929	6.183	0.296
	Residual	200,124.785	1	200,124.785
	Total	6,387,459.429	6
2.	Regression	6,166,663.393	4	1,541,665.848	13.965	0.068
	Residual	220,796.036	2	110,398.018
	Total	6,387,459.429	6
3	Regression	5,913,638.410	3	1,971,212.803	12.481	0.034
	Residual	473,821.018	3	157,940.339
	Total	6,387,459.429	6
4	Regression	5,380,981.333	2	2,690,490.667	10.693	0.025
	Residual	1,006,478.095	4	251,619.524
	Total	6,387,459.429	6
5	Regression	4,983,377.650	1	4,983,377.650	17.746	0.008
	Residual	1,404,081.778	5	280,816.356
	Total	6,387,459.429	6

and

Table 10. Linear model coefficients for UBV estimation.

Model		Unstandardized Coefficients		Standardized Coefficients	T	Sig.
		B	Std. Error	Beta
1	Constant.	615.674	605.575		1.017	0.495
	BS	−0.095	0.048	−5.710	−1.956	0.301
	BD	0.313	0.152	6.675	2.061	0.288
	BRPV	−0.094	0.086	−2.138	−1.096	0.471
	UPP	0.245	0.115	2.235	2.124	0.280
	BPM	0.570	1.775	0.141	0.321	0.802
2.	Constant.	748.404	328.958		2.275	0.151
	BS	−0.091	0.035	−5.500	−2.603	0.121
	BD	0.293	0.103	6.254	2.843	0.105
	BRPV	−0.077	0.051	−1.766	−1.514	0.269
	UPP	0.237	0.084	2.168	2.831	0.105
3	Constant.	915	370.550		2 472	0.090
	BS	1.071	0.039	−4.307	−1.836	0.164
	BD	0.198	0.098	4.223	2.025	0.136
	UPP	0.125	0.046	1.139	2.699	0.074
4	Constant.	960.083	466.718		2.057	0.109
	BD	0.020	0.016	0.425	1.257	0.277
	UPP	0.059	0.037	0.540	1. 597	0.185
5	Constant.	1308,828	396.487		3.301	0.021
	UPP	0.097	0.023	0.883	4.213	0.008

, the same conclusions can be drawn as for the UBO assessment models. Thus, the only statistically significant coefficients (Sig. < 0.05) are the coefficients of model 5.

The formed models for estimating the UBV (total number of vehicles that entered—accessed the parking facility) can be mathematically recorded in the manner presented in

Table 11. UBV estimation models—total number of vehicles that entered (accessed the parking lot).

MODEL	Mathematical Equation of the Model
Model 1	$UBV=615.674-0.095·\mathrm{B}\mathrm{S}+0.313·\mathrm{B}\mathrm{D}-0.094·\mathrm{B}\mathrm{R}\mathrm{P}\mathrm{V}+0.245·\mathrm{U}\mathrm{P}\mathrm{P}+0.570·\mathrm{B}\mathrm{P}\mathrm{M}$
Model 2	$UBV=748.404-0.091·\mathrm{B}\mathrm{S}+0.293·\mathrm{B}\mathrm{D}-0.077·\mathrm{B}\mathrm{R}\mathrm{P}\mathrm{V}+0.237·\mathrm{U}\mathrm{P}\mathrm{P}$
Model 3	$UBV=915.878-0.071·\mathrm{B}\mathrm{S}+0.198·\mathrm{B}\mathrm{D}+0.125·\mathrm{U}\mathrm{P}\mathrm{P}$
Model 4	$UBV=960.083+0.020·\mathrm{B}\mathrm{D}+0.059·\mathrm{U}\mathrm{P}\mathrm{P}$
Model 5	$UBV=1308.828+0.097·\mathrm{U}\mathrm{P}\mathrm{P}$

.

5.2. Model Testing

Model testing is performed to confirm their validity. Model testing was performed on a sample collected at LOC_3 (City Mall—Bijeljina), as a randomly selected shopping centre location. The data collected at this location were not taken into account when forming the model. A comparative view of the actual measured values of the daily number of visitors (UBO) and the number of vehicles arriving in the shopping centre zone (UBV) and the values obtained by models, for the facility at the testing site LOC_3 (City Mall—Bijeljina) is shown in

Table 12. Comparative representation of measured and values of UBO and UBV obtained by linear models for the test facility LOC_3 (City Mall—Bijeljina).

MODEL	Values Obtained by the Model		Measured Values
	Number of Visitors (UBO)	Daily Vehicle Number (UBV)	Number of Visitors (UBO)	Daily Vehicle Number (UBV)
Model 1	1259	696
Model 2	1810	988
Model 3	2691	1759	3857	2473
Model 4	3578	2316
Model 5	3746	2410

.

The results of testing the model at LOC_3 are graphically shown in the appropriate graphs (Figure 3 and Figure 4), which makes it easier to see the error size, i.e., the deviation of linear models from the measured level (represented by a dashed line).

By means of statistical analysis and assessment of the performance of the model at the LOC_3 (City Mall—Bijeljina) test site, it has been concluded that Model 5 gives the values of STABS of 3746 people and UBV 2410 vehicles, which represents the values closest to the measured ones. In this case, the all-day deviation from the measured value for UBO is 111 persons, and for UBV 63 vehicles. Taking into account the values for hourly flow distribution during the day, the differences in the sizes of the determined and calculated relevant values of the UBO and UBV at the observed position in the peak hour amount to 10 persons and 6 vehicles, which can be accepted as a relative error given the character of the traffic flow. That is, the presented Model 5 largely describes the actual measured values (

Table 13. Comparative view of the measured values of UBO and UBV and obtained by the established linear Model 5 for the test facility LOC_3 (City Mall—Bijeljina).

MODEL	Values Obtained by the Model		Measured Values of
	Number of Visitors (UBO)	Daily Vehicle Number (UBV)	Number of Visitors (UBO)	Daily Vehicle Number (UBV)
Model 5	3746	2410	3857	2473

).

Based on all the results shown above, it has been concluded that the best results at the test location for both variables are shown by Model 5, which has the following forms of finite equations:

• Model for predicting the total daily number of people on arrival—both drivers and passengers, who use vehicles to visit the mall (UBO):

  $$UBO=1963.657+0.157·UPP$$
• Model for predicting the total daily number of vehicles in the shopping centre zone accessing the parking facility (UBV):

$$UBV=1308.828+0.097·UPP$$

where: $UPP$—Total sales area.

Restrictions for the listed models:

• Models can be applied in. cases where the total sales area of the shopping centre (UPP) has a value between 4000 and 30,200 m².

Table 14. Comparative presentation of measured values and values obtained by the established model—Model 5 (UBO and UBV) at other locations.

Facility	Measured Values of		Values Obtained by the Model
	Number of Visitors (UBO)	Daily Vehicle Number (UBV)	Number of Visitors (UBO)	Daily Vehicle Number (UBV)
Object LOC_1	6639	4159	6705	4238
Object LOC_2	6283	4030	5182	3297
Object LOC_4	4799	3053	5418	3443
Object LOC_5	2637	1589.	3345	2162
Object LOC_6	3049	1972	2592	1697
Object LOC_7	3858	2513	3282	2124
Object LOC_8	2831	1936	3612	2327

shows the actual values of the daily number of visitors and the number of vehicles and the values obtained by the established model for the facilities used in the formation of the model. As the model is formed on the basis of data recorded at the locations of the mentioned facilities, the calculated values cannot be displayed as model test results, but the given values can be seen as an indicative estimate of the expected error for other objects with similar characteristics.

A summary comparative view of all measured and UBO and UBV values obtained through all five analyzed linear models is shown in

Table 15. Comparative overview of measured and UBO and UBV values obtained through all five analyzed models.

MODEL		FACILITY
		Loc.	Loc.	Loc.	Loc.	Loc.	Loc.	Loc.	Loc.
UBO	Model 1	6636	6027	1259	5019	2569.	3416	3417	2833
	Model 2	6580	6168	1810	4939	2587	3349	3240	3063
	Model 3	6832	6106	2691	4759	3049	3327	3115	2979
	Model 4	6356	5896	3578	5431	3506	2814	3065	2975
	Model 5	6705	5182	3746.	5418	3345	2592	3282	3612
UBV	Model 1	4115	3806	696	3159	1516	2199	2198	1925
	Model 2	4199	3985	988	3226	1587	2215	2166	2052
	Model 3	4421	3935	1759	3068	1993	2199	2058	1978
	Model 4	4042	3724	2316	3458	2261	1831	2000	1957
	Model 5	4238	3297	2410	3443	2162	1697	2124	2327
Measured UBO		6639	6283	3857	4799	2637	3049	3858	2831
Measured UBV		4159	4030	2473	3053	1589.	1972	2513	1936

.

In addition to the previously presented and adopted Model 5 for predicting the total daily number of persons on arrival—both drivers and passengers who visit the shopping centre (UBO) with vehicles and for predicting the total daily number of vehicles in the shopping centre zone that access the parking lot of the facility (UBV)—it can be observed that Model 4 also makes deviations that are in the domain of relative error (Table 12, Figure 3 and Figure 4).

Model 4 has the following forms of finite equations:

• Model for predicting the total daily number of people on arrival—both drivers and passengers, who use vehicles to visit the shopping centre (UBO):

  $$UBO=1367.230+0.034·BD+0.092·UPP$$
• Model for predicting the total daily number of vehicles in the shopping centre zone accessing the parking facility (UBV):

$$UBV=960.083+0.020·BD+0.059·UPP$$

where

• $BD$—Number of households.
• $UPP$—Total sales area.

Comparative view of measured values and values of UBO and UBV obtained by Model 4 for the test facility LOC_3 has been shown in

Table 16. Comparative view of measured values and values of UBO and UBV obtained by Model 4 for the test facility LOC_3 (City Mall—Bijeljina).

MODEL	Values Obtained by the Model		Measured Values of
	Number of Visitors (UBO)	Daily Vehicle Number (UBV)	Number of Visitors (UBO)	Daily Vehicle Number (UBV)
Model 4	3578	2316	3857	2473

, while comparative presentation of measured values and values obtained with Model 4 (UBO and UBV) at other locations has been shown in

Table 17. Comparative presentation of measured values and values obtained with Model 4 (UBO and UBV) at other locations.

Facility	Measured Values of		Values Obtained by the Model
	Number of Visitors (UBO)	Daily Vehicle Number (UBV)	Number of Visitors (UBO)	Daily Vehicle Number (UBV)
Object LOC_1	6639	4159	6356	4042
Object LOC_2	6283	4030	5896	3724
Object LOC_4	4799	3053	5431	3458
Object LOC_5	2637	1589	3506	2261
Object LOC_6	3049	1972	2814	1831
Object LOC_7	3858	2513	3065	2000
Object LOC_8	2831	1936	2975	1957

.

Model 4 is not accepted because the constant (Sig. = 0.112) and coefficients with the variable UPP (Sig. = 0.158) and with the variable BD (Sig. = 0.210) at UBO, as well as the constant (Sig. = 0.109) and coefficients with the variable UPP (Sig. = 0.185) and with the variable BD (Sig. = 0.277) at UBV, come out of the given accuracy threshold (Sig. < 0.05)

In practical application, Model 4’s values could be taken into account and viewed as an indicative estimate of the expected error for other objects with similar characteristics, which can be seen in Table 16 and Table 17. In this case, the limitations for the above model would be that it is possible to apply it in cases where the total sales area of the shopping centre (UPP) has a value between 4000 and 30,200 m² and for towns of a size between 18,862 and 77,717 households (BD).

6. Conclusions

As part of the work, a model was formed to determine the volume of traffic generated by shopping centres in medium-sized towns in two countries of the Southeast Europe region (the Republic of Serbia and Bosnia and Herzegovina). The construction of shopping centres arrived in our region late and changed the habits of customers. It has previously been known that land use change generates new trips, so shopping centres as constant all-day zones of attraction are especially interesting for analysis in terms of their impact on changes in traffic conditions on surrounding roads.

In the first part of the paper, through the procedures of analysis and synthesis, the current knowledge about the connections between the purpose of areas with the spatial and temporal distribution of travel was considered. Despite the fact that any change in land use at a location creates the need for adequate analysis, definition of requirements and sizing of the transport network, it was concluded that there are no models in our region on the basis of which an objective assessment of the size could be performed, as well as spatial and temporal distribution of the generated newly created traffic. Shopping centres, as objects of high attraction throughout the day, require special attention.

The second part of the paper presents a survey conducted at eight different locations throughout the territory of Bosnia and Herzegovina and the Republic of Serbia. The survey included the complete working hours of all shopping centres on the representative day, which enabled the formation of a database on the size of the generated traffic. After the formation of the database, the mutual relations and connections between the size of traffic attracted by shopping centres were analyzed depending on the characteristics of shopping centres and the characteristics of the environment in which they are located. Based on the conducted analyses, models based on linear regression analysis were defined to estimate the total daily number of shopping centre users and the expected total daily number of vehicles accessing the shopping centre parking lot. The analysis explored several different aspects of trip generation, including the relationship between trip generation and combinations of several independent variables. The conclusion is that the total sales area of shopping centres is the most significant for predicting the generated customer journeys. The results of the model were tested at a randomly selected shopping centre location, where research was also conducted, but data from this location were not taken into account when forming the model. In this way, the test results showed the validity and reliability of the formed models to estimate the average daily number of shopping centre users and the average number of vehicles appearing in the shopping centre parking lot.

The obtained model is applicable in the procedures of planning and operational analyses when selecting optimal locations for the construction of new shopping centres in medium-sized towns, i.e., locations with the least negative impact on the deterioration of traffic conditions. In the Republic of Serbia, Bosnia and Herzegovina, as well as in other countries in the wider region, on a daily basis there is an increasing number of shopping centres whose activities have a significant influence on travel patterns. Thus, the model can be both applied in determining the rates of travel attractiveness for future shopping centres, and, indirectly, for a wider traffic system, channelization and control of the traffic near shopping centres, improvement of traffic infrastructure, and ultimately it plays an important part in everyday life as a support to future growth and development. Knowledge of the rate of the attractiveness of travel to shopping centres contributes to conducting an adequate policy of sustainable transport, investment, urban planning, and ultimately sustainable urban mobility.

The directions of further research should be aimed at improving the model for estimating the number of visitors and the number of vehicles visiting shopping centres analyzing other influential factors such as the following:

• The impact of public transport and cycling infrastructure,
• The location of the shopping centre in relation to the town centre due to the possibility of reaching the shopping centre on foot, micro-mobility means of transport or bicycle,
• The impact of the content and offer of the shopping centre,
• Impact of previously built shopping centres in the area of impact, etc.
• Also, directions for future research can be modeling and analyzing similar traffic activities and processes using various methods like MCDM [52,53], neural networks [54], deep learning [55], etc.