DefinePK

Abstract

These days, many cities have traffic congestion, which is a major problem during certain peak hours and causes inhabitants to experience increased stress, noise, and pollution. Neural networks (NN) along with machine-learning (ML) techniques perform effectively with large parameterized datasets and changing behaviors which causes their adoption for real-world problem solutions over traditional analytical and statistic the developed ML and DL algorithms conduct traffic flow prediction at intersections to serve as a basis for adaptive traffic control at intersections. Two approaches of evolving traffic control could be remote control of traffic light timing or an algorithm based on expected traffic levels that alters the timing. The research develops a comprehensive approach using ML and DL models which enhances prediction accuracy of road traffic flows. The proposed methodology utilizes four models: XGB Regressor, Voting Regressor, CNN-GRU and CNN-LSTM-GRU. The dataset is preprocessed through exploratory data analysis and standardized before training. The XGB Regressor model together with Voting Regressor delivered exceptional predictive results reaching R² values of 0.9406 and 0.9412 respectively. Among the models tested CNN-GRU-LSTM delivered the peak forecasting results when tested against the CNN-GRU by having an R² score of 0.9713 along with lowest MSE 0.0291 and RMSE 0.1707 and MAE 0.1105. This proved its superiority for traffic flow prediction. The improvements are significant over existing models, including MLP-NN and Stochastic Gradient, which indicates that the proposed ML-DL method can perform effectively for real time management and prediction of traffic.