Machine Learning-Based Public Transit Ridership Forecasting System

Public bus transit systems require accurate demand forecasting to enhance operational efficiency and resource allocation. This paper presents a machine learning–based framework for predicting hourly ridership at Bangalore Metropolitan Transport Corporation (BMTC) bus stops using six months of historical data collected from 26 stops across four major routes. A SARIMAX(2,1,2) × (0,1,0,24) model is employed to capture short-term autocorrelations and daily seasonal patterns. The model achieves a mean absolute error of 18.92 passengers and a root mean squared error of 23.45 on a held-out test set. To support practical deployment, the forecasting engine is integrated into a full-stack web application built with React, Node.js Express, and a Python backend. The system provides interactive visualization, route-level filtering, and real-time predictions with confidence estimates, achieving sub-second interface responsiveness and 15–30 s inference latency. The results demonstrate the effectiveness of seasonal time-series modeling for transit demand prediction and its potential to inform fleet planning and congestion management.