My thesis focuses on trustworthy artificial intelligence and, more specifically, on developing and evaluating risk mitigation methods in supervised learning. As AI systems increasingly shape decisions, recommendations, and user experiences, they raise important concerns related to harm prevention, fairness, and explicability. This thesis addresses these challenges through a combination of technical and human-centered studies. It first examines whether collaborative filtering algorithms can contribute to filter bubbles in recommender systems. It then develops and evaluates fairness-aware methods for supervised classification, including in-processing and post-processing approaches designed to reduce demographic disparities while preserving predictive performance. Beyond technical evaluation, the thesis also studies how people react to accuracy- versus fairness-oriented algorithms in decision-making contexts. Finally, it explores explainable AI for well-being prediction and assesses which explanation formats are most satisfying to users. Overall, this work contributes to the development of AI systems that are not only accurate, but also more responsible, fair, and understandable.