| Abstrakt: |
Prediction on motorcyclist severity is always a critical task for transportation system and a promising research topic in road safety studies. Machine learning models have gained popularity in the recent years due to their strong prediction accuracy. Therefore, we aim at comparing the predictive performance, including prediction accuracy and estimation of variable importance, among the machine learning models. In this study, crash data from Malaysia is used to predict the motorcyclist severity using variables such as road type, speed limit, location type and collision type. The analysis begins with the use of random forest (RF) to adequately select important features for prediction. Then, three most often used machine learning models, which are multinomial logistic regression (MLR), decision tree (DT) and support vector machine (SVM), are applied and their performances are evaluated. The results indicated that the most important features in predicting the motorcyclist severity are the number of drivers killed, and environmental factors such as traffic system, collision type and light condition. Among the three models used in this study, SVM has shown better performance with 82.14% accuracy than DT and LR. [ABSTRACT FROM AUTHOR] |
