Review on Implementation of Road Crash Injury Severity Prediction Techniques

Severity prediction of road crash injury is an important part of making traffic safer and minimizing the effect of accidents. This review paper explores various methodologies applied in predicting crash injury severity, focusing on the latest improvements in machine learning, deep learning, and statistical methods. It covers traditional approaches such as Bayesian Networks and logistic regression, alongside modern machine learning models like Graph Neural Networks (GNNs), Random Forest, Support Vector Machines (SVM), and ensemble methods. Additionally, deep learning architectures, including Convolutional Neural Networks (CNNs) and transformers, are examined for their role in improving predictive accuracy. The paper also highlights models that address challenges like unobserved heterogeneity, spatio-temporal correlations, and data imbalance, with a focus on real-time applications, feature importance, and explainability to enhance decision-making in traffic safety management. Traffic accidents are influenced by multiple factors, some of which are more significant than others, so understanding the severity of accidents is essential. Research should focus on predicting the severity of injuries in road accidents to help minimize their impact on individuals and healthcare systems. As more detailed crash data becomes available, advanced methods from machine learning and statistics are being used to better predict how severe crash injuries will be. The paper also looks at challenges like hidden factors, patterns over time and space, unbalanced data, and making models easier to understand. It compares different methods, showing their strengths and weaknesses, and gives ideas for future research on predicting crash injury severity.