Design and implementation of an artificial intelligence supported general risk management system within the scope of occupational health and safety

This thesis presents the development of an artificial intelligence (AI)-based risk management model capable of evaluating physical, chemical, biological, ergonomic, and psychosocial hazards across various sectors within the domain of Occupational Health and Safety (OHS). The proposed model aims to overcome the limitations of traditional risk assessment methods by delivering a more dynamic, interpretable, and context-specific decision support mechanism. Trained on a simulated dataset, the model employs ensemble tree-based algorithms, notably the XGBoost classifier, which achieved a high classification performance with an accuracy of 88% and an AUC score of 0.92. Complementarily, the Random Forest algorithm provided comparable accuracy while enhancing model interpretability through feature importance analysis. Beyond categorical risk classification, the model generates continuous and numerical probability scores for each observation, enabling a more precise and prioritized risk evaluation. Training and hyperparameter optimization processes were conducted on the Google Colab platform, resulting in a 3 to 6-fold acceleration compared to local systems. Furthermore, the model architecture incorporates cron-based scripting for periodic updates, ensuring adaptability and sustainability in dynamic workplace environments. Compared to classical methods such as Fine–Kinney and the 5×5 risk matrix, the AI-driven model offers higher-resolution, data-driven analyses that more transparently elucidate risk causality and enhance the efficacy of preventive strategies. Designed as a hybrid system that complements rather than conflicts with conventional approaches, the model maintains regulatory compliance while introducing foresight, flexibility, and sensitivity to OHS management processes. In summary, the developed AI-supported OHS risk assessment model, characterized by interpretability, flexibility, and sectoral adaptability, contributes to digital transformation initiatives as a learnable and sustainable decision support system.