Predicting Estimated Time of Arrival Using Boosting Models

Estimating the time of arrival (ETA) in public transportation can be challenging due to incomplete data and the complex nature of the urban environment. This study aims to address persistent criticism of the poor punctuality problem in Malaysian buses through the modeling of bus arrival time predictions. The study uses geographical and time data to predict bus arrival times through several boosting models. The data cleaning method enhanced data quality by eliminating invariable entries, segmenting the bus route for a more granular analysis, and encoding the data for improved structure and reliability. Through the implementation of Boruta for feature selection, relevant variables crucial for prediction were identified, contributing to the model's precision. The results highlighted LightGBM's superiority over AdaBoost and XGBoost, exhibiting the highest accuracy and a balanced level of complexity. This integrated methodology not only presents a robust prediction model but also showcases a potential practical implementation.