Machine Learning-based Modeling of Pedestrian Crashes: A Corridor-level Approach

Abstract

Road crashes cause significant fatalities and injuries, especially for vulnerable road users (VRUs) including pedestrians, cyclists, two-wheelers and three-wheelers. Between 2018 and 2022, the state of Kerala, India witnessed over 5,100 pedestrian fatalities which constituted 26.5% of all road crash fatalities. The present paper uses a corridor-based approach to develop pedestrian crash prediction models for Kerala. The scope of the study is limited to pedestrian crashes, including fatal, grievous injury, and minor injury crashes reported between 2018 and 2022 in the Thrissur district of Kerala. Using GIS analysis, a total of 57 vulnerable corridors were demarcated in the study area including 13 corridors on National Highways, 30 on State Highways, and 14 on Major District Roads.  Based on the literature review, 19 parameters related to infrastructure, environment, and pedestrian facilities are identified for pedestrian crash prediction. The respective values of these parameters were collected through detailed road inventory. Machine learning algorithms like Decision Trees, Random Forest, Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost) were subsequently used to predict the crashes in the vulnerable corridors. The significant parameters affecting pedestrian crashes were identified in each model through feature importance analysis. These parameters include the availability of sidewalks and crosswalks, pedestrian movement type, gender, and age. The models were evaluated based on performance metrics including accuracy, precision, recall, and F1 score. The results showed that the SVM algorithm had higher accuracy compared to the other three models.