Scholarlink

<p>This study explores the efficacy of advanced machine learning (ML) algorithms—Classification and Regression Tree (CART), Random Forest Classifier (RFC), Extra Tree Classifier (ETC), and Combined Kernel and Tree Boosting (KT-Boost)—to examine the relationship between variables related to road traffic crash risk and injury severity. To address the challenge of imbalanced data, the Synthetic Minority Oversampling Technique (SMOTE) and its variants (SMOTEENN and SMOTE Tomek) were employed. The analysis yielded significant findings after the application of these techniques. The results demonstrated that the KT-Boost algorithm on SMOTE-treated data outperformed the other classifiers, achieving a balanced accuracy of 67.75%, weighted precision of 74%, weighted recall of 71%, G-mean of 0.67, and Matthew’s Correlation Coefficient (MCC) of 0.33. These findings confirm that the KT-Boost algorithm effectively analyzed and predicted accident injury severity with high accuracy. Furthermore, the SHAP analysis revealed that road user gender, occupant age, road profile, and number of junctions were the key factors influencing accident likelihood, with females and younger individuals being more susceptible to accidents. The insights gained from this study offer significant contributions to traffic safety and provide actionable recommendations for policymakers. The proposed KT-Boost algorithm and its integration with SHAP for interpretability may be of interest to researchers focusing on modern applications aimed at improving transportation safety and injury prevention strategies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Novel Bayesian Optimized-Combined Kernel & Tree Boost Approach for Road Traffic Crash Severity Analysis

  • 期刊论文

  • 发表日期：

    2025年4月17日

查看全文
  • Kamran Aziz,
  • Feng Chen,
  • Afaq Khattak

摘要

This study explores the efficacy of advanced machine learning (ML) algorithms—Classification and Regression Tree (CART), Random Forest Classifier (RFC), Extra Tree Classifier (ETC), and Combined Kernel and Tree Boosting (KT-Boost)—to examine the relationship between variables related to road traffic crash risk and injury severity. To address the challenge of imbalanced data, the Synthetic Minority Oversampling Technique (SMOTE) and its variants (SMOTEENN and SMOTE Tomek) were employed. The analysis yielded significant findings after the application of these techniques. The results demonstrated that the KT-Boost algorithm on SMOTE-treated data outperformed the other classifiers, achieving a balanced accuracy of 67.75%, weighted precision of 74%, weighted recall of 71%, G-mean of 0.67, and Matthew’s Correlation Coefficient (MCC) of 0.33. These findings confirm that the KT-Boost algorithm effectively analyzed and predicted accident injury severity with high accuracy. Furthermore, the SHAP analysis revealed that road user gender, occupant age, road profile, and number of junctions were the key factors influencing accident likelihood, with females and younger individuals being more susceptible to accidents. The insights gained from this study offer significant contributions to traffic safety and provide actionable recommendations for policymakers. The proposed KT-Boost algorithm and its integration with SHAP for interpretability may be of interest to researchers focusing on modern applications aimed at improving transportation safety and injury prevention strategies.