Risk Analysis in Artificial Intelligence Systems: an Evaluation of a Hybrid FMEA-ML Approach for Failure Assessment in Machine Learning Models

This study presents a risk analysis approach for artificial intelligence systems using a hybrid methodology that combines FMEA (Failure Mode and Effects Analysis) with Machine Learning techniques. The proposed methodology aims to identify and mitigate potential risks in AI systems, considering both technical and operational aspects. The developed framework integrates qualitative and quantitative analysis, enabling more accurate assessment of failure modes and their impacts. Results demonstrate that the hybrid approach offers greater precision in identifying critical risks compared to traditional methods. Validation was performed through case studies in different application domains, showing the versatility and effectiveness of the proposed methodology. The contributions include a new risk analysis framework, specific metrics for AI systems, and practical implementation guidelines. The findings suggest that FMEA-ML integration can significantly improve reliability and safety of artificial intelligence systems in critical environments, providing a robust tool for risk management.