Real-time Incident Detection in Public Bus Systems Using Machine Learning.

Traffic incident detection (ID) systems use various sensors and communication technologies to identify, classify and locate traffic incidents in real time. For public bus systems, installing physical sensors such as cameras or radars is expensive. An alternative approach is to rely solely on GPS signals acquired in real-time from the bus fleet. In this work, we propose a machine learning approach for traffic incident detection trained using historical bus GPS data. The model detects incidents earlier, allowing public bus authorities and operators to mitigate the consequences of incidents. We also propose a methodology to generate incident labels from historical data based on sudden headway changes that involve multiple buses. We evaluate the model with traditional incident detection evaluation metrics: Detection Rate (DR), False Alarm Rate (FAR), and Mean Time to Detect (MTTD). The ML model achieved a DR of 79% for incidents lasting more than ten minutes, with an MTTD of 6.4 minutes and a FAR of 2.5%.