Explainable and Ethical ML for Knowledge Discovery from Medical Data Sources

The proposed framework will be based on three pillars:

1. Data integration and indexing: Since the data originates from several medical data sources, it is essential to find suitable unified representations that combine the data in a suitable way. The data sets are typically very large and have a complex structure. Hence, data indexing methods are needed for searching the data sets efficiently.
2. Explainable machine learning: Within health applications, it is fundamentally important to be able to explain the predictions provided by machine learning classifiers. Besides fairness, explainability of predictions facilitates trust in the system.
3. Legal aspects: In order to ensure ethical integrity it is necessary to detect and prevent unintended bias within the predictive models.