A Context-Aware Nonnegative Matrix Factorization Framework for Traffic Accident Risk Estimation via Heterogeneous Data

Traffic accidents have significantly globally increased over the past decades. The safety of transportation system has become an important issue for human society. Efficiently estimating accident risk will help for alleviating these safety issues and improving safety investment. As accidents are always caused by complex factors, heterogeneous data and a suitable model to combine these data information are needed in accident risk analysis. In this paper, we proposed a framework which utilizes matrix factorization method to estimate accident risk. First, we collect heterogeneous data and extract features from them so that we can get feature matrices to describe the background when accidents happened. Furthermore, we utilize context-aware non-negative matrix factorization method to model accident risk in a citywide scale. The results validate the efficiency of our model, and suggest that accident risk estimation can be significantly more accurate with heterogeneous data even accident data is missing or environment changes.