An optimized hybrid approach for trajectory prediction and vehicle classification at intersections using Dijkstra and random forest enhanced by a genetic algorithm

Given the ongoing growth in car numbers, traffic jams at urban intersections provide a major difficulty in contemporary society. Bad intersection management usually causes major delays, greater greenhouse gas emissions, and higher risk of accidents. Effective traffic management techniques meant to maximize traffic flow, improve road safety, and support environmental sustainability must be implemented in order to solve these problems. From this point of view, hybrid methods that mix optimization and machine learning seem to be good ways to model how drivers behave and plan trips efficiently. This article suggests a hybrid approach perfect for vehicle categorization at metropolitan intersections and navigation. Using a real dataset created from VISSIM simulation software, the viability of the approach is shown in a complex intersection scenario. To faithfully replicate the circumstances of a road segment, the data include several features including road segment length, lane count, and traffic capacity. The approach uses machine learning methods, Random Forest, XGBoost, and GA-RandomForest, to forecast the most appropriate vehicle type along this path, hence combining the Dijkstra algorithm for best route choice. Among these models, GA-RandomForest performs best with 95.84% accuracy, 94% precision, 98% recall, and 95% F1 score. These findings show that the suggested method has the capacity to enhance dynamic traffic management at challenging junctions as well as intelligent navigation.