Score-Level Fusion Approach for Traffic Flow Prediction in Intelligent Transportation System

Traffic flow prediction (TFP) for intelligent transportation system is essential for effective transportation planning and city development. Conventional models often have difficulty with the complex nature of traffic data, which includes nonlinear patterns, spatial dependencies, and external influences such as weather or road conditions. Prediction accuracy may be significantly increased by utilizing an ensembling strategy, which combines the capabilities of many models. The goal of the study is to improve TFP accuracy by integrating the results of different machine learning (ML) models. The study uses Random Forest(RF) Naïve Bayes(NB), K-Nearest Neighbor(KNN), Extreme Gradient Boosting(XGBoost) and Support vector machine(SVM) as prediction models and based on the prediction probabilities and the score level fusion is performed with RF and XGBoost ensembling. The new fusion procedure uses a weighted average technique in which weights are assigned dynamically depending on individual model performance criteria including accuracy, precision, recall, and F1-score. Higher-performing algorithms are given greater significance in the final forecast, resulting in a dynamic and data-driven assembly strategy. The findings show that the efficiency of the ensemble of RF-XGBoost is higher compared to the contrasted methods. The model also uses feature selection approaches like mutual information to find the most significant traffic-related characteristics, lowering computational cost and increasing forecast efficiency.