Abstract Occupational accidents remain one of the key challenges for modern work systems, both from the perspective of employee health and life, and in terms of economic, social, and organisational losses for society. Despite the significant diagnostic value of near misses, data on such events have not been incorporated as input variables in predictive models. This study aimed to develop, experimentally test, and empirically validate the effectiveness of various mathematical models in predicting occupational accidents. The performance of thirteen machine learning and deep learning algorithms, including Random Forest, logistic regression, SVC, ARIMA, SARIMAX, and a proposed modified convolutional neural network incorporating Bayesian correction and a tailored ReLU activation function, was compared. The best results were achieved with a CNNKT model, yielding an accuracy of 0.939, an F1-score of 0.934, and a G-Mean of 0.942. The findings confirm the superiority of deep convolutional networks over traditional statistical models in occupational safety analysis.
A 2026 study studied this question.