Comparative Analysis of Machine Learning Algorithms with Feature Selection for Predicting AADT

Abstract: Accurate prediction of Annual Average Daily Traffic (AADT) is crucial for effective transportation management, infrastructure planning, and policymaking. This study investigates the application of machine learning models to enhance AADT prediction by selecting the most influential features. Three feature selection methods—Filter, Wrapper, and Embedded—to identify key features and evaluate their impact on model performance were employed. This comprehensive analysis demonstrated that the filter method significantly improved prediction accuracy, with the Random Forest model exhibiting the best performance among the evaluated models. Specifically, the filter method with Random Forest achieved the lowest Mean Absolute Error (MAE) and Mean Squared Error (MSE) of 0.4883 and 0.4253, respectively. Along with the highest R-squared (R²) and Explained Variance Score (EVS) of 0.7537 and 0.7538, respectively, indicating a superior fit and predictive power. Key features identified as crucial for AADT prediction include the number of lanes, speed limit, percent of grade, and degree of curve. These features were found to have the most significant impact on traffic volume, highlighting their importance in developing robust predictive models. These findings have important implications for transportation engineering and planning, providing a data-driven approach to improve traffic volume predictions. By leveraging effective feature selection techniques, transportation agencies can make more informed decisions regarding infrastructure development, traffic management, and safety improvements. This research underscores the critical role of feature selection in enhancing the accuracy and reliability of machine learning models for AADT prediction.