Back to search

Democritus University of Thrace (DUTH)

Πρόβλεψη των μελανών σημείων του οδικού δικτύου Μακεδονίας και Θράκης με συμβατικές μεθόδους και με μεθόδους μηχανικής μάθησης

Abstract

dc:description

The present thesis titled "Prediction of black spots of the road network of Macedonia and Thrace with conventional and machine learning methods", focuses on the comparative evaluation of conventional and innovative methods of black spots prediction in a road network. By providing a publicly accessible database, called Black Spots of Northern Greece (BSNG), and a highly accurate method of detecting road black spots, it offers a potential solution to challenges related to the identification and analysis of the latter, such as the quality of data, their limited availability, the difficulty in accurately assessing the factors that contribute to road accidents as well as the continuous evolution of road design, infrastructure, and vehicle safety systems. Road black spots are also referred to as black spots, hot spots, accident-prone locations, sites with promise and hazardous road locations. They are spatiotemporal phenomena and refer to specific parts of the road network where the frequency and consequences of road accidents are significant over some time. Assessing these problematic sites helps researchers uncover the underlying causes and reasons for increased crash rates, such as road design, traffic volume, driver behavior, weather conditions, and infrastructure. The identification of black spots involves the analysis of the geographical location and the timing of road accidents. Typically, this analysis focuses on specific locations on road networks during defined periods in order to identify areas with a higher concentration of accidents, known as black spots. It is one of the most important steps in understanding the factors that contribute to road traffic accidents and in developing strategies to reduce the number and severity of these incidents. Furthermore, it is vital to prioritize road safety interventions and evaluate their effectiveness in reducing the number of road accidents. The thesis is developed in three (3) sections: The first section entitled "OVERVIEW OF THE ROAD SAFETY ISSUE" is structured in ten (10) chapters. In particular, Chapter 1, entitled "ROAD TRAFFIC ACCIDENTS", attempts a first approach to road traffic accidents. The latter, as a factor of the morbidity and mortality of the population in most countries worldwide and a constant threat to road users, are among the list of the most important global social problems. They are complex phenomena, caused by multiple interconnected factors, with humans at the top of the list. They have a huge economic impact, amounting to 518 billion dollars worldwide, but they also have a psychological impact in terms of anxiety, severe post-traumatic stress and depression. Road accidents, as shown in Table 1, are the 1st cause of death for road users aged 15-29 years and the 8th cause of death for all ages. Without appropriate measures, it is predicted that by 2030 it will rise to 5th place. The same chapter stresses the importance of analyzing and investigating them in the context of integrated and effective road safety management, along with their definition, the presentation of their evolution and the development of the three categories (fatal, injury, and property damage) into which they are divided based on their effects. Chapter 2, entitled "ROAD SAFETY", begins with a definition of road safety. It then describes the three factors to which it is causally related, which are none other than the road user, the vehicle and finally the road and the road environment. Table 3 shows the participation percentages of the above factors involved in the road accident. The human factor is at the top with 65%. The literature review revealed that: 1) The road user with his/her individual elements and behaviors such as age, gender, alcohol consumption, mobile phone use, speeding, distraction, etc., plays a key role in causing road accidents. 2) Vehicle characteristics such as age, size, weight, mechanical failures due to age or poor maintenance, etc., are important factors of road accidents. Figure 2 shows the relationship between the age of the vehicle and its mechanical damage. 3) The design and construction of road infrastructure have a significant impact on road safety. Road characteristics such as roadway width, curves, limited visibility, road surface type, etc., have an impact on road accident rates. In addition, adverse environmental conditions (e.g. snow, rain, fog), as well as inadequate or complete lack of road lighting, have an impact on the occurrence of road accidents. Chapter 3, entitled "ASSOCIATION OF ACCIDENTS WITH DRIVERCHARACTERISTICS", describes the individual driver characteristics associated with an increased risk of being involved in a road accident, such as age, gender, and experience. Regarding age, the Chapter states that drivers under 25 and over 65 years of age are the highest risk group for road accidents. Some of the factors attributed to the increased risk of young drivers being involved in road accidents are their impulsiveness, overestimation of their driving skills, lack of experience and tendency to engage in anti-social and dangerous behavior. In the case of older drivers, this risk is due to the normal loss of capacity associated with mental, ophthalmological, and otological conditions. About gender, the conclusions of the various studies converge on the fact that female drivers are less at risk of being involved in a road accident than male drivers. Finally, experience is in turn a factor that influences road accident rates. According to various studies, higher driving experience may imply an increase in the ability to judge dangerous situations more accurately and avoid them, but also an increase in confidence and overestimation of abilities that result in high-risk behaviors. Therefore, the acquisition of experience, which is largely dependent on the annual number of driving kilometers, the number of years of active driving and previous involvement in an accident that affects driving behavior, can either reduce or increase the number of accidents. Chapter 4, entitled "'ASSOCIATION OF ACCIDENTS WITH VEHICLE CHARACTERISTICS", shows how age (seniority), size and type of vehicle are associated with the causation of road accidents. The age of the vehicle was found to be directly correlated with the number of fatalities per 100 million vehicle kilometers of vehicles of the same age (Figure 3). For the size of the vehicle, which is characterized by its weight and constitutes characteristic inextricably linked to road safety, it is reported that its reduction results in an increase in fatalities (Table 5), as well as in moderate and serious injuries. Finally, the type of vehicle has a strong influence on the safety of drivers, passengers, and pedestrians, as well as on the severity of an accident. In particular, motorcycles, which are vehicles with higher risk characteristics, have higher road accident rates (Tables 6 and 7). Chapter 5, entitled "RELATIONSHIP OF ACCIDENTS WITH ROAD AND ENVIRONMENTAL CHARACTERISTICS", analyses the impact of the geometric and traffic characteristics of the road, as well as the various weather conditions such as snow, fog, rain and temperature, on the occurrence of road accidents. The geometric characteristics of the road include its cross-section and horizontal and vertical alignment. These are factors that affect road safety. Their improvement can significantly reduce the frequency of road accidents. In the case of the cross-section, reducing the width of the lane and the width of the curb, which are among its main parameters, drastically increases the likelihood of accidents. Regarding the horizontal curve, its specific characteristics that influence the occurrence of road accidents are its radius, curvature, length, and frequency. Analyses have shown that the smaller the radius of the curve, the higher the accident rates (Table 8). The same applies for the other two characteristics of the horizontal curve. That is, as the length of the horizontal curve and the frequency of curves on a road section increases, the accident rates increase (Table 9). Another geometric feature of the road is the vertical alignment. The vertical curve is an important factor in road accidents. Its danger depends on its inclination. Its increase leads to an increase in the accident index (Table 11). The influence of the specific slope on the risk of the curve depends on the radius of the curve (Table 12). As for the traffic characteristics of the road, it can be seen that understanding them is necessary to improve traffic management and reduce the frequency of accidents. These characteristics include traffic volume and speed. Traffic volume, which is related to the traffic capacity of the road, is sometimes a major factor in causing road accidents and at other times is not the main factor or has no influence. It is a measure of exposure and is used in the study and analysis of road accidents. Speed as a traffic parameter influences the probability, frequency, and severity of accidents. Its influence on the accident index is exponential (Figure 6). In general, its relationship with road accidents can only be interpreted if the simultaneous influence of other road elements related to traffic flow, environment, and vehicle is taken into account. Finally, environmental conditions appear to be a critical element that should be taken seriously in the context of managing and dealing with road accidents. Rainfall and limited visibility due to fog, snow, and temperature are common weather parameters. Rainfall has been identified as the most important weather parameter in road safety studies. Its impact aggregates the effects of risk and exposure. Most studies have shown that there is a positive correlation between rainfall and collision frequency. The same positive correlation is observed in the case of fog and snow, weather phenomena that contribute to an increase in both the number of road accidents and their severity. Finally, temperature, which reflects seasonal weather conditions, cannot be described as a major cause of road accidents. Researchers have reached conflicting conclusions as to the type (positive or negative) of its correlation with the frequency of road accidents. Chapter 6, entitled "DAY-NIGHT", shows how conditions of day or darkness affect the occurrence of road accidents. Day and night are two variables that express travel time. They have a significant influence on the number of road accidents and their severity. In Chapter 7, entitled "ROAD LIGHTING", road lighting, the lack of which is inextricably linked to the causing of road accidents, is presented as an important driving safety factor. Chapter 8, entitled "ROAD SAFETY AND TRAFFIC ACCIDENTS", firstly describes the three (3) parameters that influence road safety, namely exposure to risk expressed through a set of human behaviors that expose the individual to risk, the risk assumed by the user when undertaking a dangerous action and the consequences of the road user's actions, which describe the impact of the road accident in terms of property damage and human lives. All of the above is described with the help of the road safety cube (Figure 9). Then, in light of the new traffic conditions and the large number of deaths caused by road accidents at both national and European levels, the need to adapt and improve the level of road safety is emphasized. The application of benchmarking practice is considered by several countries to be the most appropriate way of achieving this improvement. Chapter 9, entitled "ROAD SAFETY IN GREECE", presents the fluctuations in road safety in Greece since the 1990s.During this period, deaths and road accidents have been on the increase due to several factors, such as the lack of infrastructure, the limited funding for road safety, the inefficient organization of public administration, the increase in the number of vehicles in circulation and the lack of traffic education up to the present day. The analysis of the data of road accidents that occurred during the years 2014-2018 and the recording of the main road safety indicators show a decrease in the number of road accidents in the year 2018 compared to the year 2014 (Table 13), which as a percentage change reaches 8.152% (Table 14). There is a similar trend in terms of deaths, which decreased from 795 in the year 2014 to 700 in the year 2018 (Table 16). The male gender shows a decrease in deaths of 13.2% followed by the female gender with a decrease of 6.1% (Table 19). In terms of age groups, the largest decrease in deaths is in the 0-24 age group at a rate of 27.9%, while the only group that shows an increase, at a rate of 19.7%, is the age group of 65 years and over (Table 20). As for the number of injured, during the above period, it is decreasing (Table 17). Although the above converges to the conclusion that the level of road safety in Greece has improved significantly in recent years, unfortunately, this improvement cannot be considered satisfactory compared to the performance of other developed countries. In Greece, during the period 2001-2020, three (3) strategic plans, each with a separate structure and priority axes, were developed in order to improve the level of road safety. In this context, programs were developed and targets were set for implementation. The above effort is complemented by several organizations dealing with road safety issues, such as the Hellenic Statistical Authority, the Hellenic Police with its competent services, the Hellenic Institute of Transport, etc. In the 10th chapter entitled "ROAD SAFETY AT EUROPEAN AND INTERNATIONAL LEVEL", the level of road safety in various countries of the world is first presented with the help of road safety indicators. According to these indicators, the Old Continent has the lowest road accident mortality rate in the world (Figure 28). The level of road safety in its countries varies (Table 24), reflecting different cultures, demographic factors, modes of transport, etc. During the years 2014-2018, the number of road accidents in the European Union is decreasing (Table 23). In the same period, Greece, a European Union member state, showed the largest decrease in the number of deaths due to road accidents (Figure 30). Finally, the stages of the implemented European Union policy to improve road safety are analyzed and the European Union-level bodies dealing with road safety issues are mentioned. The second section, entitled "ANALYSIS OF THE ROAD SAFETY ISSUE - IMPLEMENTATION FIELD OF APPLICATION", is structured into six (6) chapters. More specifically, Chapter 11 describes the procedure to be followed for collecting and processing the required data and the individual stages of processing the data required. More specifically, the Hellenic Statistical Authority provided upon request the required data referring to road accidents that took place on the national and provincial network of Macedonia and Thrace during the years 2014-2018 and resulted in the injury or death of some of the persons involved in them. In total, six (6) excel files were provided containing variables related to the vehicle, the victims, and the road accident along with characteristics of the road environment. After merging the excel files, creating new data files, removing incomplete and unclear data, as well as other data to ensure uniformity of the mileage locations under consideration, classifying and further anonymizing them, the final number of 1811 road accidents was obtained. Reference is also made to a set of assumptions, which refer to the adoption of a procedure for the extraction of traffic loads, to the difficulties related to the application of the methods chosen for the identification of black spots and to the limitation of the reduction of the data processed, which would make it impossible to analyze them further. Chapter 12 defines what a road black spot is. The importance of identifying and analyzing black spots to prevent road accidents and thus promote road safety is stressed. The methods for identifying black spots are listed. These are distinguished into: 1) classical methods including numerical (accident frequency method, gravity index method) and statistical methods (quality control method, Poisson random distribution method), 2) analysis method based on geographic information system (GIS) technologies used for mapping road accidents, 3) accident reconstruction method involving the use of various techniques, such as simulation models, to recreate the conditions leading to the road accident, 4) road safety audit method which involves a thorough examination of the road network and the road environment, with an emphasis on identifying infrastructure and road design issues that may contribute to road accidents, and 5) deep learning-based methods. The following is a summary of the criteria used by different countries to identify black spots (Table 27). Finally, four (4) classical methods of identifying black spots (Poisson Random Distribution Method, Quality Control Method, Gravity Index Method, Accident Frequency Method) are applied at 90% significance level, the coupling of which results in the number of black spots (142 spots) for the road network under consideration in Macedonia and Thrace. Chapter 13 presents the independent variables (ordinal, binary, categorical, numerical) and the dependent dichotomous variable (Table 29) that were used to create our model using the Statistical Package for Social Sciences (SPSS). Reference is made to how some of the variables managed to obtain their final form, and the reasons for not using some of them. By way of example, the variable "road gradient" was obtained by combining the variables "downhill" and "uphill", while the variable "road way condition" was not used, since in 99 % of the cases under consideration it is characterized as normal. Chapter 14 provides a theoretical analysis of the concept of regression. Due to the inclusion of more than one independent variable in our model and their strong correlation, which leads to the occurrence of multilinearity, the ways of identifying and solving it are discussed. Then, the concept of logistic regression is developed, as well as a category of it, namely bivariate logistic regression. The methods (enter, forward, backward) of automated selection of the optimal model using the statistical program SPSS are mentioned, as well as the disadvantages that determine their choice or not. At the theoretical level, various statistical criteria (e.g., z criterion), measures of goodness of fit (e.g., Cox and Snell's R2 coefficient of determination) and hypothesis evaluation (e.g., Chi-square test) associated with the logistic regression model are presented. Chapter 15 refers to the data that was discussed in the previous chapter. Thus, first, a linear regression analysis is applied using SPPS. From this, through the VIF and Tolerance values of the Coefficients table, the dimensions of the Collinearity Diagnostics table and the dimension values of the status indicators, it is found that the variables "LEFT EDGE LINE" and "RIGHT EDGE LINE" exhibit collinearity. The same applies for the variables "STRAIGHT", "RIGHT TURN", and "LEFT TURN". To remove the collinearity among the solutions, we decided to remove all five variables from the model. Subsequently, a logistic regression analysis is applied using the Enter and Backward LR methods. Difference values of indicators and measures of the aforementioned analysis are interpreted and the final parsimonious model, obtained using the Backward LR method, is formulated in equation form. Chapter 16 describes the capabilities of machine learning algorithms (artificial neural network, support vector machine, and deep learning models) in the field of identifying black spots in a road network. Their ability, in the case of handling complex and highly non-linear data, to overcome the limitations and weaknesses of applied statistical modeling is highlighted. Some general principles of machine learning are mentioned and ten (10) well-known supervised learning algorithms are described. Subsequently, the results of applying ten (10) classical machine learning methods, logistic regression using SPSS, and the method proposed in this thesis are presented. The performances (Accuracy, Precision, Recall, F1-score, AUC) of the above methods are compared with the data: 1) in their original form, 2) encoded using the one-hot technique and 3) augmented through the MixUp technique (Table 31). The above comparison shows that the proposed method using a shallow MLP (multilayer perceptron) and the low dimension of the latent features obtained via MixUp outperforms all other methods in all metrics. Finally, an evaluation of the results is given. In terms of this evaluation, the following are mentioned: Random Forest and XtraTrees performed better than AdaBoost in detecting accidents that occurred in black spots. When increasing the sample population and using the coding technique of the proposed method performance gains were achieved. The proposed method outperformed all other methods in terms of Accuracy, Recall, F1-score, AUC, and Precision. This suggests that it may be a promising solution to the black spot detection effort, despite the challenges posed by the unbalanced and complex "Black Spots of Northern Greece" dataset. The results of the present research show that the impact of traffic volume on the identification of black spots is still uncertain, low visibility is a factor affecting road safety, the non-availability of a valid technical inspection certificate may not have a significant impact on the occurrence of accidents, and the age group most frequently involved in road accidents is not young or novice drivers but people driving cars for more than 10 years. The above findings challenge previously established perceptions about the main causes of road accidents. The 3rd section entitled "CONCLUSIONS OF THE RESEARCH - LIMITATIONS OF THE RESULTS - CONTRIBUTION OF THE DISSERTATION TO THE PROMOTION OF ROAD SAFETY", consists of one chapter. Chapter 17 contains the main conclusions of the research concerning road safety, the limitations in the dissemination of its results, and finally the contribution to the field of the research. As far as the conclusions are concerned, it is stated that the identification of black spots is a complex and dynamic field that requires the integration of various data sources and advanced analytical techniques. This study proposes a four-step approach as a possible solution to the challenges related to black spot detection, such as the quality and limited availability of data and the difficulty in accurately assessing the factors contributing to road accidents. Regarding the limitations, it is mentioned that the study focused exclusively on the regions of northern Greece and the specific time frame of 2014-2018. Therefore, the findings may not represent traffic types and characteristics (infrastructure, enforcement policies) of other regions of Greece or different periods. Finally, this thesis aims to provide an overview of the different methods for predicting black spots in the road network and their strengths and limitations. Moreover, it discusses their possible applications in the field of road safety. Finally, its contribution is to provide an in-depth literature review, the publication of a data set for the identification of black spots on the provincial and national road network of Macedonia and Thrace (BSNG) and a highly accurate method for the prediction of the latter, as well as the establishment of a benchmark for the identification of problem areas called black spots on Greek road networks.

Degree

thesis:*
Grantor dc:publisher
Democritus University of Thrace (DUTH)
Year dc:date
2024

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Karamanlis, Ioannis
  • Καραμανλής, Ιωάννης

Subjects

dc:subject × 14

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/56788

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Karamanlis, Ioannis; Καραμανλής, Ιωάννης. Πρόβλεψη των μελανών σημείων του οδικού δικτύου Μακεδονίας και Θράκης με συμβατικές μεθόδους και με μεθόδους μηχανικής μάθησης. Democritus University of Thrace (DUTH), 2024. http://hdl.handle.net/10442/hedi/56788