[3Xin4] Poster session 1

[3Xin4-32] Traffic Risk Mining Method Considering The Scale Of Damage

〇Ryo KUWAMOTO¹, Daisuke NAKAI¹, Yuto TAKAKAI¹, Shuhei KUWATA¹, Daichi MOCHIHASHI² (1.Mitsui Sumitomo Insurance Co.,Ltd., 2.The Institute of Statistical Mathematics)

Keywords:Traffic Risk Mining, Non-Negative Matrix Factorization, Traffic Accident Data

We propose a method that predicts whether the traffic accident happens at any given location by using traffic accident history data. In particular, our method can take into account the magnitude of damage and simultaneously identify factors that cause accidents to occur. Therefore, for example, we can alert drivers when they are approaching an area where there is a high risk of accidents, even in places where no accidents have occurred before. The key points of our method are that the alert level can be changed according to the severity of the accident and that countermeasures derived from the factors leading to the accident can be presented when the alert occurs. The proposed method predicts the insurance amount as an approximation of the severity of the accident. By using actual accident data, we compare our method with conventional method based on non-negative matrix factorization and evaluate the characteristics and usefulness of the method.