JJEM: Special Edition 2 (December 2024)
JJEM: Special Edition 2 (December 2024)
2024-12-08
Machine learning based predictions of traffic accident severeness
Kavana V H, Adarsh M J
This article, examined several factors including their effects on accident extent prognosis. For us to do such, researchers examined a number of overall techniques employed in machine learning (ML), both single and ensemble mode, and evaluated how well they performed regarding area under the receiver operator characteristic (AUROC), know, efficiency, and precision. In order to categorize the intensity of an accident into two groups, (i) a binary rating (like grievous and non-grievous), and (ii) multiclass classification (catastrophic, significant, small, along with noninjury), the road crash degree prognosis problems was taken into consideration in this study. In comparison to other methods considered in this research, our findings show that Random Forest (RF) outperformed them in both single and ensemble ML methods, such as logistic regression (LR), K-nearest neighbour (KNN), naive Bayes (NB), extreme gradient boosting and adaptive augmentation (e.g., 86.64% for binary and 67.67% for multiclass classification). Single-mode machine learning techniques like LR, KNN, and NB perform comparably in binary and multiclass classification. With the order of RF, XGBoost, and Adaboost, ensemble machine-learning techniques can forecast the accident severity more precisely than single-mode methods can. The results of this investigation can be used to learn more about the factors that led to the accident and the extent of injuries sustained as a result.
Random forests, decision trees, logistic regression, and machine learning