IMPROVING BUS BOARDING DEMAND ESTIMATION USING DEEP LEARNING INTEGRATED WITH REAL-TIME TRANSIT DATA

Abstract

Passenger demand variability poses a significant challenge for public transportation systems, with more than 60% of daily bus boardings occurring during peak hours and nearly 30% of commuters experiencing delays due to overcrowding. In dense urban transit networks, inaccurate demand estimation contributes to ineffective route optimization and inefficient resource allocation. Traditional demand estimation approaches—such as time-slot passenger counts, on-ground route surveys, and ticket ledger analysis—are often labor-intensive, error-prone, and difficult to scale, resulting in inconsistent planning decisions. To address these challenges, this study proposes a deep learning–based bus boarding demand estimation framework integrated with real-time transit data. The proposed system employs Support Vector Machine (SVM) and Deep Neural Network (DNN) classifiers trained on historical and real-time features, including timestamps, route identifiers, traffic delays, and boarding volumes. Model performance is evaluated using standard accuracy metrics and validated against conventional manual estimation methods to ensure reliability and robustness. The results demonstrate that the AI-based approach delivers high-precision, real-time demand predictions, enabling dynamic scheduling and improved alignment between passenger demand and service supply. This data-driven framework provides a scalable solution for enhancing operational efficiency and planning effectiveness in modern public transportation systems.