Intelligent and Efficient Traffic Accident Prediction and Control System using Random Forest Algorithm

Abstract

Road traffic accidents remain a major public safety challenge, necessitating intelligent prediction systems that can accurately assess accident risk under dynamic traffic and environmental conditions. This study helps to design a smart system of predicting traffic-related risks and accidents with the help of the Random Forest algorithm based on the analysis of lane traffic density, weather, and signal status. The system estimates the risk of accidents in real time to facilitate proactive safety and risk conscious traffic control. Materials and Methods: The study utilizes historical traffic datasets collected from Kaggle and Data.gov. There are Two experimental groups are considered: Group 1 uses the XGBoost algorithm with a sample size of 3,000 instances of traffics for traffic prediction and signal timing analysis. Group 2 uses the proposed Random Forest Algorithm-based prediction system with a sample size of 3,000 instances of traffic. Results: The accident risk prediction model was a proposed random forest model which attained accuracy of 95.2% including 94.9% precision, 97.2% recall and the F1- score of 96.1% which is superior to the XGBoost model that reported an accuracy of 83.5. Even though the Random Forest model had a reduced prediction error (5.1%) than XGBoost (8.4%), the difference was not statistically significant at the 95% confidence level (p = 0.3). The 4-lane accident risk prediction system proposed on the basis of the Random Forests is an accurate prediction of the probability of accidents on a lane based on the parameters of traffic and weather. It is also applicable in real time traffic safety and decision support applications because it remains stable even in dynamic conditions.