This study focuses on the application and methodical evaluation of various machine learning algorithms and deep neural networks to predict survival probabilities in traffic accidents. The research systematically implements parametric methods (Logistic Regression) alongside non-parametric approaches (including k-Nearest Neighbors, Decision Trees, and Random Forest) as well as deep learning-based Deep Neural Networks across extensive datasets. Rigorous methodological procedures such as data preprocessing, model training, hyperparameter optimization and cross-validation support the evaluation of prediction performance through metrics like accuracy, precision, recall, F1 score and the area under the ROC curve.

Analyses reveal that each algorithm exhibits a distinct capacity to capture the inherent structural and statistical characteristics of the data. In this context, optimizing predictive strategies for survival probabilities in traffic accidents requires careful consideration of factors such as computational efficiency, interpretability, and generalization ability. The findings underscore the importance of data-driven approaches in strengthening the methodological foundations of traffic safety research, thereby providing a robust reference point for future investigations in this domain.