Development of the Model for Determination of Car Crash Subcenters Using Machine Learning

This study is devoted to the development of the model for determination of car crash subcenters using machine learning. The proposed model is based on the approaches used in machine learning to predict probability of a car crash. It consists of a predictive algorithm and map visualizing crash risk subcenters, predicted by the algorithm. In terms of the study a comparative analysis of current car crash prediction methods was made providing a basis for the proposed XGBoost method. Also, an analysis of the causes of accidents was made, on the basis of which the features used by the predictive algorithm was formed. The developed predictive algorithm was tested using the implemented software on dataset containing 223594 samples. The best recall indicator is 93 %. Recommendations on the applicability of the proposed method and its software implementation are given. Scope of work: 72285 characters, 11 images, 11 tables, 44 references.