Ondersteuning van de implementatie van AI in Overlevingsduuranalyse, met gebruik van Verklaarbare AI technieken en Active Learning

Artificial Intelligence (AI) has become a crucial part of many businesses and society, with applications ranging from recommendation algorithms to generative AI. However, the lack of transparency in AI models, often seen as 'black boxes', limits their adoption in critical areas like healthcare and banking, where errors can have serious consequences. This thesis tackles the transparency issue by introducing new methodologies that bridge the gap between Explainable AI (XAI) and complex models like Survival Analysis. The contributions are threefold: First, a novel interpretability method, Bellatrex, is introduced to enhance the transparency of Random Forest and Random Survival Forest models by extracting human-readable rules while preserving accuracy. Second, an XAI pipeline for healthcare applications is developed, focusing on explainable prognoses for Multiple Sclerosis patients using Survival Analysis. Finally, a new Active Learning framework is proposed for time-to-event data, improving model training efficiency and revealing insights through effective sampling strategies. These advancements aim to make AI more accessible and reliable in high-stakes decision-making environments.