Enhancing Safety Prediction in Accident-prone Zones using Gradient Boosting Machines

The aim of this study is to improve safety prediction in accident hotspots through the use of gradient boosting machines (GBM) and comparing their performance with K-Nearest Neighbors (KNN). It measures predictive accuracy, validity, and overall performance in the analysis of accident data. Materials and methods: Group 1applies KNN for safety prediction using 26 samples, measured on 90.4% accuracy, 88.6% recall precision, and F1-score. Group 2 applies GBM for accident risk estimation using 26 samples, measured on the same performance metrics accuracy of 97.5%, a recall of 96.8%. Results: Accuracy for GBM was between 85.4% and 96.7% with a high of 92.5% at a level of significance of 0.0048, while KNN varied between 72.3% and 89.1%. Conclusion: GBM's better feature selection and ensemble learning make it a superior model for real-time estimation of accident risk.