A Novel Hybrid XGB-RF BlendStack Model for Traffic Flow Prediction

Abstract

Accurate short-term traffic flow prediction is crucial in the realm of Intelligent Transportation Systems (ITS), driving effective planning and management. Existing methodologies often falter in capturing the intricate nonlinearities within traffic flow dynamics, resulting in suboptimal prediction accuracy. Addressing this, our study introduces a novel hybrid deep learning and ensemble model harnessing Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU) networks, as well as Random Forest (RF) and XGBoost algorithms. his fusion is supposed to help automatically extract the important contents from the traffic flow features. Ensemble methods, renowned for combining diverse model strengths to bolster overall performance, have steered attention within predictive modeling. Our proposed hybrid technique unites two prominent ensemble algorithms, namely RF and XGBoost, within a stacked ensemble framework. Initially trained on a UCI Machine Learning Repository public dataset, individual Random Forest and XGBoost models merge their predictions through a stacking strategy. Subsequently, a linear regression model synthesizes insights gleaned from these base models using the stacked features. Empirical experimental results reveal that the XGBRF BlendStack model achieved a remarkable performance with a Mean Absolute Error (MAE) of 114.52, a Root Mean Square Error (RMSE) of 0.03, a Mean Absolute Percentage Error (MAPE) of 66.18%, and an (R)-squared (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>) value of 0.99. These values not only demonstrate the superiority of the XGB-RF BlendStack model over standalone RF and XGBoost models but also underscore the efficacy of our hybrid approach in significantly improving predictive accuracy for short-term traffic flow forecasting.