Occupational Diseases Risk Prediction by Neural Networks

Abstract

This study explores the use of neural networks for occupational disease risk prediction based on worker and workplace characteristics. The goal is to develop a tool to assist occupational physicians in monitoring workers. Using a dataset from the Italian MalProf National Surveillance System (2019-2023), an ensemble of one-vs-all classifiers is trained to identify six prevalent disease classes. Performance is evaluated using accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). The results indicate promising performance. The specificity values for all six disease classes under study exceed 0.920 on average over 10 runs, and for five out of six classes, they surpass 0.967. Regarding sensitivity, the performance is positive (average over 10 runs greater than 0.920) for all classes, except for “Carpal Tunnel Syndrome and other Mononeuropathies of Upper Limb”, which performs less effectively (average over 10 runs = 0.655). Future research could focus on optimizing neural network architectures, applying oversampling techniques for underrepresented classes, and analyzing misclassifications.