Impact of Key Drivers on New York Traffic Flow: A Comprehensive Study

Abstract

This study aims to predict traffic volumes in New York City using integrated datasets from transportation, weather, and collision records. The author employed multiple machine learning models, including Random Forest Regressor, XGBoost, LightGBM, and an Ensemble Model combining Random Forest and LightGBM, to utilize their strengths. The data preprocessing involved cleaning, merging, and encoding categorical features, resulting in a comprehensive dataset that integrated traffic volume, collision, and weather data. The findings indicated that the LightGBM model achieved the highest accuracy with the lowest error rates (MSE: 1336.12, RMSE: 36.55, MAE: 26.35, R²: 0.9788). While comparing with other studies, the models, particularly LightGBM, provided superior predictive performance. Despite the promising results, the limited dataset size and computational complexity posed challenges. Therefore, in the future, the research should focus on expanding the dataset, exploring advanced ensemble techniques, and evaluating the applicability of the model across different subpopulations to enhance the robustness and universality of traffic volume predictions.