Comparative Study of Road Transportation Prediction System Using MlPNN and SVM

Vehicle-related Road accidents remain one of the most distressing experiences, often leaving lasting physical, emotional, and economic consequences for victims and society. Frequent road accidents and vehicle collisions are largely attributed to poor road conditions, environmental influences, and driver negligence. Predicting accident severity can assist safety agencies and transportation professionals in understanding key contributing factors and improving preventive strategies. Machine learning techniques provide an effective means of identifying accident patterns and estimating severity levels such as fatal, major, or minor injuries. This study proposes a machine learning–based approach for predicting traffic accident severity using two algorithms: a multilayer perceptron neural network (MLPNN) and a support vector machine (SVM). Road traffic accident datasets sourced from the Kaggle repository were used for experimentation. The MLPNN model incorporated a decay-based regularization technique to constrain large weights, thereby reducing model complexity and improving generalization. Hyperparameters were carefully optimized to enhance classification performance. Experimental results showed that the MLPNN achieved a prediction accuracy of 98.88%, outperforming the SVM, which recorded an accuracy of 89.0%. Based on these findings, the MLPNN model is recommended for reliable traffic accident severity prediction and effective extraction of critical accident features.