Adaptive-XGBoost for Improved Long-Term Multi-Step Traffic Flow Prediction

Long-term multi-step traffic flow prediction is crucial for effective traffic management.Recently, XGBoost has demonstrated its capability in multi-step prediction applications.However, optimal hyperparameter tuning using adaptive intelligent methods like the Particle Swarm Optimization (PSO) algorithm has yet to be explored.This paper presents an Adaptive eXtreme Gradient Boosting (Adaptive-XGBoost) model utilising the direct method with the combination of Grid Search and PSO algorithms for adaptive hyperparameter tuning for multi-step prediction.This approach aims to demonstrate the potential of Adaptive-XGBoost compared to existing deep learning models like Long-Short Term Memory (LSTM) and Transformer in terms of Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE).Experimental results show that the Adaptive-XGBoost model achieved a 3.10% and 4.98% improvement in MAE over the Transformer and LSTM models, respectively, and a 0.26% and 5.69% improvement in RMSE.These findings highlight the potential of Adaptive-XGBoost for improved long-term multi-step traffic flow prediction.