A Methodology to Find the Importance of Winter Road Characteristics on Winter Road Accidents

Various factors such as road geometry, precipitation, freezing temperature, and ice on the road surface increase the risk of different types of road accidents in winter. This study proposes a methodology to classify and model winter road accidents and determine the importance of each input variable (locational characteristics, road characteristics, and winter weather characteristics). This methodology utilizes a machine learning method for multi-class classification and then applies an approach to identify important input variables affecting classification of winter road accidents. The methodology has seven main stages and starts with data analysis, which gives a general overview of the dataset and is a major stage in using machine learning algorithms. Next, four different classes regarding personal injuries are defined for road accidents. After dividing the dataset into training and testing sets, categorical variables need to be transformed into numerical variables to be understood by machine learning algorithms. Then, different models for multi-classification need to be trained and tested to find the model with the best performance based on various evaluation metrics and plotting the process of learning and testing the model. Finally, the recursive feature addition method can be used to rank the importance of input variables on classifying severe road accidents in winter.

Keywords: Machine learning, Multi-class classification, Winter road accidents, Personal injuries, Recursive feature addition.