Leveraging Process Mining and Deep Learning to Improve Health and Safety Outcome Predictive Models

The ability to predict future events is a valuable asset across many sectors, enhancing decision-making, risk management, and system efficiency. This thesis investigates the application of deep learning in predictive modeling across three key domains: process optimization, healthcare, and occupational safety. The first contribution introduces the Adjacency Matrix Deep Learning Prediction (AXDP) Model, which combines graph theory and deep learning to forecast sequential events. AXDP demonstrated superior performance on eight public datasets, outperforming existing models in prediction accuracy. The second contribution presents a deep learning model to predict mortality in ICU patients with Paralytic Ileus, using ablation and SHAP analyses to identify critical clinical predictors. The third contribution focuses on predicting neurologic outcomes in out-of-hospital cardiac arrest (OHCA) patients, showing that early clinical data can effectively forecast outcomes and that novel biomarkers may further improve accuracy. The final contribution explores whether occupational safety and health (OSH) professionals are prepared to integrate AI into workplace safety efforts. To address this, the Artificial Intelligence Learning Framework for Occupational Safety Professionals (ALFO) was developed and evaluated. In two studies involving 114 participants, ALFO significantly improved AI literacy, with 80% of participants recognizing AI’s potential to enhance safety. Together, these contributions highlight the impact of predictive deep learning and targeted education in advancing safety, efficiency, and healthcare outcomes.