ADAPTIVE DEEP-LEARNING FRAMEWORK FOR REAL-TIME HOURLY BUS BOARDING DEMAND CLASSIFICATION IN URBAN TRANSIT

Abstract

Public transportation systems face significant fluctuations in passenger demand, with over 60% of daily 
boardings occurring during peak hours and nearly 30% of commuters experiencing delays due to 
overcrowding. In densely networked urban transit systems, about 70% of route optimization efforts fail 
without accurate boarding predictions. Traditional manual demand estimation techniques such as time
slot passenger counts, field route surveys, and ticket ledger analysis are often hindered by human error, 
time-consuming procedures, and limited scalability, resulting in inconsistent data and ineffective 
planning. To address these challenges, this study introduces an AI-driven boarding demand prediction 
model that utilizes Support Vector Machine (SVM) and Deep Neural Network (DNN) classifiers. The 
system integrates real-time data inputs, including timestamp, route number, traffic delays, and historical 
boarding volumes, to train and evaluate the models using accuracy-based metrics for optimal prediction 
performance. Furthermore, the AI-based results are benchmarked against three conventional analysis 
methods—time-slot monitoring, on-site route surveys, and ticket data review to ensure result reliability 
and model robustness. The proposed solution delivers a high-accuracy, real-time prediction framework 
that supports dynamic scheduling and minimizes the mismatch between supply and demand. This 
integration of machine learning into public transport planning offers a scalable, data-centric alternative to 
manual approaches, paving the way for smarter, more efficient transit systems.