A context-aware evaluation method of driving behavior

As Uber-like chauffeured car services become more and more popular, many drivers have joined the market without special training. To ensure the safety and efficiency of transportation services, it is an important task to accurately evaluate the driving performance of individual driver. Most of the existing methods basically depend on the statistic of abnormal driving events extracted from individual vehicles. However, the occurrence of abnormal events can be affected by various factors, such as road conditions, time of day and weather. It can be bias to judge the driver’s performance by merely counting the abnormal events without considering the driving context. In this paper, we analyze the influence of driving context over driving behaviors and propose a context-aware evaluation method. Instead of taking all the occurrence of driving events as the same, we adopt the TF-IDF to determine the risk weight of a driving event in a specific driving context. Based on the risk-weighted statistics, we evaluate the driving performance precisely and normalize it using the Z score model. An evaluation system is implemented. We evaluate the effectiveness of our method based on a real dataset with 3-year traces of 1000 drivers. The normalized score determined by our method have a greater correlation (0.611) with the accident records than that of the number of abnormal driving events (0.523).