Classification of Spatial Anomalies in Bus Trajectories: A Machine Learning Approach

Public transportation, essential in urban and suburban settings for its environmental and operational efficiencies, is becoming increasingly sophisticated with Automated Vehicle Monitoring (AVM) technologies. These systems capture and process a wide array of data in real-time, facilitating unprecedented service monitoring and management. This research uses vast GPS (Global Positioning System) datasets from bus routes to analyze spatial anomalies in bus trajectories. Using modern AVM systems, the study applies Machine Learning techniques, specifically Random Forest and Extreme Gradient Boosting (XGBoost), to classify trips (defined as individual journeys from a specific origin to a specific destination at a given time) based on detected anomalies such as route deviations and unexpected interruptions. This classification enables targeted corrective actions, enhancing service reliability. Public transport operators in Italy and Europe receive remuneration based on various factors, including kilometers covered, and face penalties for not meeting service quality conditions. The Machine Learning tool developed to detect and classify spatial anomalies offers significant advantages in cost management, compliance with service obligations, and operational efficiency. The study outlines a methodical approach involving feature engineering on GPS points, aggregation into trip datasets, preliminary categorization of anomalies, and detailed multi-class classification using advanced algorithms. Focusing on a practical application, the study evaluates GPS data from Autoguidovie, a local operator in northern Italy. The highest-performing model, Extreme Gradient Boosting, demonstrated a classification accuracy of 88.72%. In the case study, the model identified 9,511 trips affected by spatial anomalies out of 27,832 trips analyzed, generating nearly 4 million GPS points. This result enabled the appropriate management of anomalies through proper analysis and corrective actions.