查看文章

Analysis of crash injury severity is a promising research target in highway safety studies. A better understanding of crash severity risk factors is vital for the proactive implementation of suitable countermeasures. In literature, crash injury severity was widely studied using statistical models. Though these models have a sound theoretical basis and interpretability, they were based on several unrealistic assumptions, which, if flouted, may yield biased model estimations. To overcome the limitations of statistical models, applied machine learning has rapidly emerged on the horizon of highway safety analysis. This study aims to model injury severity of motor vehicle crashes using three advanced machine learning approaches, i.e., vanilla multi-layer perceptron (MLP) using Keras, MLP with embedding layers, and TabNet. Among the three models, TabNet may be considered a fairly complex framework which is based on …