Sensing and classifying roadway obstacles: The street bump anomaly detection and decision support system

Theodora S. Brisimi, Setareh Ariafar, Yue Zhang, Christos G. Cassandras, Ioannis Ch. Paschalidis

Abstract: We develop an anomaly detection and decision support system based on data collected through the Street Bump smartphone application. The system is capable of effectively classifying roadway obstacles into predefined categories using machine learning algorithms, as well as identifying actionable ones in need of immediate attention based on a proposed “anomaly index.” We introduce appropriate regularization to the classification algorithms we employ, which has the effect of utilizing a sparse set of relevant features to perform the classification. Further, our novel “anomaly index” allows us to prioritize among actionable obstacles. Results on an actual data set provided by the City of Boston illustrate the feasibility and effectiveness of our system in practice.