Machine Learning-Assisted Vehicle Collision Detection Model (MLVCDM) for Road Intersections

The increase in the number of fatalities caused due to road accidents is one of the world’s most pressing challenges. As per studies, the number of road accidents at intersections is more than the number of accidents on other road segments. The proposed study aims to design an intelligent system using machine learning models to reduce traffic accidents’ frequency and its impact at road intersections. The proposed work presents a Machine Learning-Assisted Vehicle Collision Detection Model (MLVCDM) for road intersection collision detection. The proposed MLVCDM undergoes model training by utilizing the publicly accessible Kaggle dataset and is based on accident detection from CCTV videos. Before building a new set of features via feature engineering, the MLVCDM performs data preprocessing using the Kaggle dataset. Next, the balanced dataset has been processed using three machine learning algorithms: XGBoost, LGBoost and CatBoost. The three models have been ensembled in the proposed MLVCDM using the bagging technique and have been compared with the basic machine learning models, boosting models, and other ensembled models like stack ensemble and vote ensemble. The results reflect that the proposed MLVCDM model outperforms other models with an accuracy of 98%.