Analysis on fatal and severe bicycle accidents in the state of California

Abstract: According to the California Highway Patrol, over the ten year span from 2002 to 2012, there were more than 143,000 bicycle accidents in the state of California. In this article, we study the bicycle-involved accidents over these years in California. Our main objective is to explore and identify the relationship between various accident-contributing factors (e.g. age, location and time) and accident severity. To achieve our objective we performed a variety of frequency-based analysis, studied the application of feature ranking algorithms, applied density based clustering both regionally and statewide, and explored the use of classifiers to predict accident severity. We identified age, sex, time of day, and alcohol involvement all as attributes which affect accident severity. By using a feature ranking system, we verified a relationship between accident location, sex, and the use of safety equipment as effecting the severity of accidents. Our clustering analysis shows accident hot-spots both within the state of California and large urban cities like San Francisco and Los Angeles. The classification results demonstrate a remarkable ability to determine accident severity when given data from the current year, consistently giving accuracy above 90%. When we tried using the same classification algorithms to predict accident severity on future data, our results fell to around 50% for all classifiers. Keywords: Bicycle Accident, Severity Analysis, Clustering, Classification, Feature Ranking