Predicting the severity of an accident is undoubtedly a fundamental aspect of an accident event. Identifying the primary factors affecting the severity of road accidents is essential for minimizing the level of accident severity. Accidents can be predicted and prevented by analyzing accident events and identifying their patterns. Due to the large number of existing datasets, machine learning is taking the lead over traditional statistical approaches in Prediction. One potential way to predict accident severity is to use machine learning algorithms. Therefore, machine learning algorithms have emerged as one of the possible approaches for modeling accident severity due to their excellent results.
This research aims to implement an accident prediction framework using random forest, decision tree, naive Bayes, gradient boosting, k-nearest neighbor, and voting classification algorithms. The results obtained from running the algorithms on accident data to predict the impact of accident severity on traffic showed that the best algorithm is the random forest with an accuracy of 0.908, and the lowest accuracy (0.25) belongs to the naive Bayes algorithm. To improve the probability of predicting accident severity, the voting classifier algorithm was used with a set of two machine learning algorithms (random forest and extreme gradient boosting) due to their higher accuracy compared to other algorithms, and the other algorithms were removed from this set due to their lower accuracy. Thus, using the voting classifier algorithm, the Prediction of accident severity was improved with an accuracy of 0.91. Validation of the spatial accuracy of accident prediction resulting from the model and comparison of accurate data with prediction data resulting from the model also showed that the covariance between the two images is 0.99, indicating a high correlation between the two images.