A Voting Classifier Approach for Hourly Passenger Boarding Demand Prediction Using Synthetic Data

Smart-card data has emerged as a valuable source for analyzing passenger boarding behavior and predicting future public transit demand. However, acquiring real-world smart-card data poses challenges due to privacy and availability constraints. This study addresses this limitation by generating a synthetically simulated smart-card dataset that mimics hourly passenger boarding patterns at bus stops. One of the major challenges in such datasets is class imbalance, where the number of non-boarding instances significantly outweighs the boarding instances. To tackle this, we propose a voting-based ensemble learning approach that combines multiple classifiers to enhance prediction performance. Unlike prior research which employed Deep Generative Adversarial Networks (GANs) and Deep Neural Networks (DNNs) on real datasets, our approach focuses on classical ensemble methods to offer interpretability and computational efficiency. Experimental results on the simulated dataset demonstrate the effectiveness of the voting classifier in addressing class imbalance and predicting hourly boarding demand. This paper provides an alternative perspective on demand prediction using ensemble techniques and synthetic data, highlighting its potential for future real-world applications.