AI-based Countermeasure Recommendation Systems for Traffic Safety

Abstract: Improving traffic safety and reducing deaths and serious injuries is very important and necessary, but it is not easy to conduct improvement projects because of the large amount of time and cost. In this respect, selecting the most effective improvement method and applying it to the field is the most core and important step in the entire traffic safety improvement project process. However, despite this importance, a systematic and scientific approach was lacking in the process of selecting improvement methods. On the contrary, it has been argued that the involvement of expert’s subjective judgments reduces objectivity and makes it unclear whether the improvement is effective compared to the cost invested.

In this study, we proposed a system that could suggest the most effective method to users by introducing artificial intelligence techniques in the process of selecting traffic safety improvement countermeasures. In this study, data construction was conducted in three stages to develop the system. In the first step, the causes were analyzed based on past accident history data and constructed with actual accident cause data. In the second step, data on potential causes of accidents were constructed from areas that were selected as vulnerable points in terms of road geometry and safety facilities, although they were not areas where accidents actually occurred. The third step was to build an integrated database of an artificial intelligence-based improvement countermeasure recommendation system by linking these two aspects of data with improvement countermeasures that had a significant effect on reducing traffic accidents in terms of economic analysis.

Using the integrated database, we built an artificial intelligence system that can recommend optimal improvement countermeasures. In order to verify the results of the proposed artificial intelligence-based recommendation system, an in-depth interview survey was conducted with 100 traffic safety experts to verify the accuracy of the recommendation results. As a result of applying artificial intelligence techniques based on the learning data built through this study, it was found that superior scores were obtained compared to the previously used improvement countermeasure selection method in all items, including recommendation accuracy, economic feasibility, and expert satisfaction.