REAL-TIME PASSENGER FLOW PREDICTION USING AUTOMATED DATA PREPARATION

Abstract

Public transportation systems are vital for sustainable urban development, addressing critical challenges such as traffic congestion, environmental impact, and urban mobility. However, the dynamic and unpredictable nature of passenger demand complicates resource allocation and operational efficiency. Traditional forecasting methods, including ARIMA, fail to account for the non-linear and spatio-temporal complexities inherent in modern transit systems, necessitating the adoption of more advanced methodologies. This study introduces an innovative real-time passenger flow prediction framework tailored for Astana’s public transport network. The framework leverages state-of-the-art machine learning models, including LightGBM, Random Forest, and conceptual Spatio-Temporal Graph Neural Networks (ST-GNNs) proposed for future implementation. While the ST-GNN component is outlined conceptually, the experimental evaluation in this study focuses on ensemble and boosting models such as LightGBM and Gradient Boosting Regressor. A comprehensive data preprocessing pipeline was developed to address common issues such as missing values, outliers, and temporal inconsistencies, ensuring high-quality input data for predictive modeling. The proposed framework achieved a Mean Absolute Error (MAE) of 1.06, a Root Mean Squared Error (RMSE) of 1.70, and a Mean Absolute Percentage Error (MAPE) of 34.8%, demonstrating superior predictive performance compared to traditional methods. This accuracy enables precise demand forecasting, optimized resource allocation, and enhanced passenger experiences. Unlike traditional approaches, the framework integrates advanced spatio-temporal modeling techniques, bridging the gap between theoretical advancements and practical applications. This research offers actionable insights for the development of intelligent transportation systems, providing a scalable and adaptive solution to the challenges of urban mobility in rapidly growing cities.