Artificial Intelligence-Based Model for Automated Assessment and Prediction of Occupational Risks

The studies are reviewed under five topics: weaknesses of traditional risk assessment methods, risk level classification, forecasting incidents over time, explaining model decisions (SHAP), and analysing the text of incident reports. The review finds that ensemble methods are most often used for tabular data, while LSTM networks can perform well in short-term forecasting. However, assessment and forecasting are usually solved separately, models are rarely tested at another enterprise, and the small number of severe incidents keeps class imbalance a problem. Taking these shortcomings into account, a conceptual model with data preparation, classification and forecasting modules is proposed. The model has not yet been tested in practice and serves as a basis for further research.