Machine Learning and Visualization of Sudden Braking using Probe Data

This paper presents a novel mining and visualizing tool that detects features to estimate sudden braking. The tool uses a machine learning and feature selection method to find the features exhaustively from combinations of the features which include not only vehicle-related factors, but also outer circumstances or temporal factors. The tool also obtains the locations inferred by the features detected. A normal way would first search for locations where sudden braking behavior frequently occurred, but it is not always true that the occurrence probability of sudden braking at the locations is high. On the other hand, our tool finds the locations related to sudden braking with high probability, more than 98%. Through the visualizing process, the features can be used as clues to find new factors which affect sudden braking.