رژين زارعي

Urban traffic flow management an‎d delay reduction in transportation networks represent major challenges in metropolitan areas. Considering the complexity an‎d dynamic nature of traffic flows, employing advanced machine learning an‎d deep learning techniques offers an effective approach to enhancing the efficiency of urban transportation systems. In this research, real-world traffic data were utilized an‎d preprocessed through han‎dling missing values, normalization, an‎d origin–destination matrix construction. Subsequently, a set of machine learning models, including Ran‎dom Forest an‎d XGBoost, along with deep learning architectures such as LSTM, Attention-based LSTM (A-LSTM), an‎d Convolutional LSTM (ConvLSTM), were developed an‎d eva‎luated. The results revealed that the ConvLSTM model equipped with an attention mechanism achieved the highest predictive performance, with a coefficient of determination R²=0.9331. The A-LSTM (R²=0.9290) an‎d baseline LSTM (R²=0.8929) models ranked next, whereas tree-based models exhibited inferior accuracy due to their inability to capture temporal dependencies. The principal contribution of this thesis lies in the design an‎d implementation of a hybrid ConvLSTM framework augmented with an attention mechanism, enabling simultaneous spatiotemporal feature extraction an‎d substantially improving predictive accuracy. The findings provide a robust foundation for the development of intelligent traffic control systems an‎d evidence-based policymaking in the domain of urban transportation management.

يادگيري ماشين , جريان ترافيك شهري , شبكه حمل و نقل

machine learning , urban traffic flow , transportation network

Rojin Zarei

Dr Shahriar Afandizadeh Zargari