Effectiveness of trip planner data in predicting short-term bus ridership

Predictions on Public Transport (PT) ridership are beneficial as they allow for sufficient and cost-efficient deployment of vehicles. On an operational level, this relates to short-term predictions with lead times less than hour. Where conventional data sources ridership, such Automatic Fare Collection (AFC) data, may have longer lag contain no travel intentions, in contrast, trip planner often available (near) real-time used before traveling. In paper, we investigate how from a app can be utilized bus predictions. This is combined AFC (in case smart card data) construct ground truth actual ridership. Using informative variables the dataset through correlation analysis, develop 3 supervised Machine Learning (ML) models, including k-nearest neighbors, random forest, gradient boosting. The best-performing model relies forest regression requests. Compared baseline that depends weekly trend, it reduces mean absolute error by approximately half. Moreover, using same without prove usefulness improved 8.9% 21.7% increased coefficient determination 5-fold cross-validation 7.8% 18.5% two study lines, respectively. Lastly, show performance maintained even requests prediction up 30 min ahead, different periods day. We expect our methodology useful PT operators elevate their daily operations level service well companies facilitate passenger replanning, particular during peak hours.