Snow Removal Dispatch Prediction Using Road Surface Image and Weather Information with L1 Regularization

Abstract

Snow removal from roads in snowy areas during the winter is essential for maintaining road traffic and infrastructure. However, since the decision to dispatch snow removal operations is greatly influenced by changes in weather and road conditions, the decision made by the person in charge is often reversed, which not only forces the snow removal crew to prepare for uncertain dispatch, but also places a psychological burden on the person in charge. To solve this problem, we propose a snow removal dispatch prediction method that integrates road surface images with diverse weather information, utilizing L1 regularization for feature selection. Specifically, we automatically select effective features for model learning from a large set of candidates including both qualitative and quantitative data such as hourly temperature, snowfall, and wind direction thereby enabling high precision predictions. Our experimental results show that the proposed method achieves higher accuracy compared with both manual judgment by experienced personnel and a conventional logistic regression model.