SMARTRIDENET: DEEP LEARNING FRAMEWORK FOR IMBALANCED SMART-CARD DATA–DRIVEN BUS DEMAND PREDICTION

Abstract

Efficient urban mobility requires accurate forecasting of passenger boarding demand to ensure optimal resource allocation and service planning in public transportation systems. Traditional forecasting models struggle with imbalanced smart-card datasets, where peak-hour records dominate while off-peak data remain underrepresented, resulting in biased and unreliable predictions. This paper introduces SmartRideNet, a deep learning framework designed to predict hourly bus boarding demand using imbalanced smart-card transaction data. The model employs a Long Short-Term Memory (LSTM) architecture combined with Synthetic Minority Over-sampling Technique (SMOTE) and focal loss optimization to address imbalance issues and capture temporal dependencies effectively. Extensive experiments conducted on real-world smart-card datasets demonstrate that SmartRideNet outperforms existing machine learning approaches such as Random Forest, ARIMA, and Gradient Boosting in both accuracy and robustness. The proposed framework provides real-time, reliable, and interpretable demand forecasting, enabling smart transportation authorities to optimize fleet operations, reduce passenger waiting times, and support data-driven urban transit planning