KNOWLEDGE BASED PREDICTION SYSTEM FOR ROAD TRAFFIC CONGESTION

Abstract

Traffic congestion has become a ubiquitous issue in urban areas worldwide, posing significant challenges to transportation systems, economies, and quality of life. In the context of India, where rapid urbanization and population growth exacerbate the problem, addressing traffic congestion is of paramount importance. This thesis introduces a comprehensive approach to tackle road traffic congestion by developing a knowledge-based prediction system tailored to the Indian scenario in context to inter-urban highways combining user and expert perspective as well as data driven approach to prepare knowledge base for traffic congestion. This thesis represents the integration of two primary modules: one is founded on user survey methodologies, while the other is centred on the application of deep learning techniques for understanding the congestion dynamics. The objectives of this thesis encompass various aspects of congestion prediction. Firstly, it aims to identify congestion attributes from travellers’ perspective, understanding the factors contributing to congestion through observation and feedback. This objective is performed by considering data from the 13 different locations situated along the undivided inter urban highway sections screened out on the basis of mixed land use pattern, where the problem of congestion is recurring in nature. The data has been collected using the survey questionnaire with a total response of 206 travellers. For the identifying the heterogeneity existence among the two traveller groups (drivers and passengers) Mann Whitney U test has been applied. Other than congestion attributes identification this objective ranks these attributes using the Relative Importance Index (RII) and rank their preference on strategies to reduce it also. And proposes possible suggestion on the traveller’s-based policy recommendations for improving there travelling experience on Inter Urban Highways. Secondly, thesis second objective explores the interrelationship among congestion factors using expert knowledge and applying exploratory factor analysis on the attributes obtained from objective one. This objective has used 282 responses and 5 expert’s input as data sample. These experts belong to different experiences in transportation sector. The modelling of interrelationship between factors has been obtained by using the fuzzy based total interpretive structural modelling technique (Fuzzy TISM) that gives the direct and indirect links of cause and effect between the congestion factors obtained after performing the exploratory factor analysis on the traffic congestion attributes identified in objective one using interpretive knowledge base provided by the experts. To identify the independent, dependent, linkage and autonomous factors in the interrelationship digraph obtained from Fuzzy TISM, Micmac analysis has been applied. Results shows that exploratory factors analysis highlights the eight significant factors of congestion on inter urban highways namely road geometry deficiencies (C1), Environmental factors(C2), External events(C3), Archaic traffic management(C4), Inefficient Public transport (C5), Driving behaviour(C6), Special events (C7) and Regional Economic Dynamics(C8). The application of Fuzzy TISM technique segregates factors into different level of the hierarchical interrelationship diagraph. Where, Level I factor are C3, C4 and C6. Level II factor is C5, Level III factor are C1 and C7 and IV factor are C2 and C8. The conclusion that can be drawn from the fuzzy Micmac analysis are factors like Special events (C7), Environmental Factors(C2) and Regional Economics Dynamics (C8) are the independent factors that influences most of the factors even though they have less driving power, and occurrence of these factors also impacts the linkage factor. Archaic traffic management (C4) is the Linkage factor in this study, which has both high driving and high dependence power due to its unstable property it impacts and get affected by the other factors too like C1(Road geometry deficiencies), C3 (External events), C5 (Inefficient public transport facilities) and C6 (Driving behaviour). First two objectives complete the first module of this thesis. The second module of the thesis contains objective three and objective four. Thirdly, the study aims to automate the process of determining traffic volume using advanced deep learning detectors and trackers, enabling real-time data acquisition for congestion prediction. This objective has been performed on the custom dataset of 10,000 Indian vehicle images. This dataset includes the images of various backgrounds scenes from highways containing multi vehicles in images. To develop the model for vehicle classification and traffic volume counting the model has been trained using single stage object detection technique called YOLO (You Only Look Once) variations and Byte tracker for counting. The result of the traffic volume counter is analysed on the intersection scenes as well as mid blocks of the inter urban highways on the videos varying in length of 1 min, 2 mins and 5 mins, which gives satisfactory results on the short min videos with an accuracy of 98 % and for long videos with an accuracy up to 87%.The fourth objective of the study proposes a knowledge base prediction system framework for road congestion that combines the results of the objective one and objective three by integrating impedance factors identified in the first objective and traffic flow variables obtained from the third objective to construct a comprehensive knowledge base for predicting congestion state. This objective has been performed on the data sample of 84 hours for weekdays and weekends. To analyse and generate the if then rules for decision makers ANFIS (Adaptive Network based fuzzy inference system) technique has been applied based on different membership functions. It shows the variations in the congestion levels on weekdays and weekends are due to the presence of multiple impedance factors. Lastly, the proposed knowledge base road traffic congestion prediction system leverages both user-driven insights and expert knowledge to enhance the accuracy and effectiveness of congestion prediction. By combining user perspectives with expert analysis and automated data acquisition techniques, the system offers a holistic approach to understanding and forecasting traffic congestion. This comprehensive methodology is particularly relevant in the Indian context, where diverse traffic conditions and complex urban landscapes necessitate adaptable and accurate prediction systems. Also, the integration of fuzzy logic and deep learning techniques enables the system to handle the inherent uncertainty and complexity of real-world traffic scenarios, providing robust and reliable predictions. Moreover, by incorporating user feedback and expert insights, the system can continuously adapt and improve its predictive capabilities, ensuring its relevance and effectiveness in dynamic traffic environments. In conclusion, the proposed knowledge base road traffic congestion prediction system offers a promising solution to the pervasive problem of traffic congestion in India. By leveraging a combination of user perspectives, expert knowledge, and advanced technologies, the system provides a comprehensive framework for understanding, analysing, and predicting congestion levels. Implementation of this system has the potential to significantly improve traffic management strategies, alleviate congestion, and enhance the overall efficiency of urban transportation systems in India.