GPS-Based Bus Stop Prediction Using Machine Learning: A Data-Driven Approach for Smart Urban Transit

Abstract - Urbanization and increased reliance on public transport warrants efficient bus stop identifying. This research utilizedmachine learning and DBSCAN clustering to predict bus stops from GPS traces (GPX files) identifying true stops versus temporary stops. Key steps included preforming preprocessing of GPX data (latitude, longitude, speed, timestamp), detecting zero-speed periods, and clustering stop points based on frequency and confidence. The model for predicting bus stops scores GPX files for support and frequency of stops that occur in a particular GPX file. The resulting system tested against OpenStreetMap (OSM) data and manual surveys accurately identified bus stopsin a variety of urban settings, including urban settings with density and complexity. Utilization of interactive maps and statistical visualizations supported better interpretation of identified bus stops, and the overall approach is easily scalable for transit agencies to better plan transit in their context and hopefully improve passenger experience across cities and contexts.