Sorum Neero Gumsar, Sorum Martina Gumsar
Department of Civil Engineering, North Eastern Regional Institute of Science & Technology, Nirjuli, Arunachal Pradesh, India.
PLoS One. 2025 Jun 16;20(6):e0326113. doi: 10.1371/journal.pone.0326113. eCollection 2025.
Distracted Driving (DD) is one of the global causes of high mortality and fatality in road traffic accidents. The increase in the number of distracted driving accidents (DDAs) is one of the concerns among transportation communities. The present study aimed to examine the individual and interacted effects of the influential factors on the injury severity of the DDAs using the Binary Logistic Regression (BLR) method, and at the same, to select the best machine learning (ML) model in predicting the injury severity of the DDA. The selection of the best ML model was based on the optimum combination of accuracy, F1 score, and area under curve metrics. Ten years of DDA data (2011-2020) provided by the police department of Imphal, India, was used in the present study. The BLR model-without-interaction results revealed that out of twenty categorical variables, nine categorical variables (below 18, 18-24, 25-40, above 40 years age group, two-wheeler, heavy motor vehicle, 12AM-6AM, 6PM-12AM, and hit-object collision) were statistically significant to the injury severity of the DDAs. In interaction model results, there were 11, 1, and 1 significant combinations among categorical variables in two-way, three-way, and four-way interaction models, respectively. The ML model results showed that overall, the XGBoost model was reported as the best-performing model in the first hyperparameter set, and the Single Layer Perceptron model in the second set. These results may be useful for transportation policymakers while implementing any countermeasures to improve road safety in hilly areas.
分心驾驶（DD）是导致道路交通事故高死亡率和高致死率的全球性原因之一。分心驾驶事故（DDA）数量的增加是交通领域关注的问题之一。本研究旨在使用二元逻辑回归（BLR）方法检验影响因素对DDA伤害严重程度的个体和交互作用，并同时选择预测DDA伤害严重程度的最佳机器学习（ML）模型。最佳ML模型的选择基于准确率、F1分数和曲线下面积指标的最佳组合。本研究使用了印度英帕尔警察局提供的十年DDA数据（2011 - 2020年）。无交互作用的BLR模型结果显示，在二十个分类变量中，九个分类变量（18岁以下、18 - 24岁、25 - 40岁、40岁以上年龄组、两轮车、重型机动车、午夜12点至凌晨6点、下午6点至午夜12点以及碰撞物体）对DDA的伤害严重程度具有统计学意义。在交互作用模型结果中，双向、三向和四向交互作用模型的分类变量之间分别有11个、1个和1个显著组合。ML模型结果表明，总体而言，在第一组超参数中，XGBoost模型被报告为表现最佳的模型，在第二组中则是单层感知器模型。这些结果可能对交通政策制定者在实施任何改善山区道路安全的对策时有用。
