Application of Machine Learning and Deep Learning Approaches for Traffic Operation and Safety Assessment at Signalized Intersections

The exponential traffic growth hasn\u27t been well-handled by traditional control systems. Adaptive controls are necessary at signalized intersections since they foresee traffic demand based on AI approaches and make decisions ahead of time. These approaches also boost traffic safety by predicting near-crash events leveraging cutting-edge datasets like LiDAR. This thesis addresses such applications of machine learning and deep learning approaches using emerging traffic datasets. A novel deep learning model, MGCNN is suggested for short-term turning volume prediction using GRIDSMART data from the MLK corridor in Chattanooga, Tennessee. During assessments for 1-to-5-minute future prediction, MGCNN surpasses contemporary models with 0.9 MSE. Traditional machine learning models are applied efficiently for forecasting speed and arrival at green with 0.04 and 0.05 MSE. Convolutional Gated Recurrent Neural Network model is proposed for near-crashes prediction that shows 100% recall, precision, and F1-score: accurately predicting all near-crashes based on LiDAR data from Georgia intersection, MLK