Utilizing Machine Learning Models to Predict the Car Crash Injury Severity among Elderly Drivers
Rabia Emhamed Al Mamlook, Tiba Zaki Abdulhameed, Raed Abdullah Hasan, Hasnaa Imad Al-Shaikhli, Ihab Mohammed, Shadha Tabatabai
Abstract
Car crash can cause serious and severe injuries that impact people every day. Those injuries could be especially damaging for elderly drivers of age 60 or more. The goal of this research is to investigate the risk factors that contribute to crash injury severity among elderly drivers. This is accomplished by designing accurate machine learning based predictive models. Naïve Bayesian (NB), Decision Tree (DT), Logistic Regression (LR), Light-GBM, and Random Forest (RF) model are proposed. A set of influential factors are selected to build the five predictive models to classify the severity of injuries as severe injury or non-severe injury. Michigan traffic data of the elderly population is used in this paper. Data normalization and Synthetic Minority Oversampling Technique (SMOTE) as injury classes balancing technique are used in the pre-processing phase. Results show that the Light-GBM achieved the highest accuracy among the five tested models with 87%. According to the Light-GBM model, the three most important factors that impact the severity of injuries are the driver's age, traffic volume, and car's age.