Analysis of Explainability ML Techinques against Adversarial Models

We want to test some local and global explainability techniques on classic ML models, such as LIME, DiCE, and SHAP. These tests will be useful to verify the fairness of the trained models and the importance of each feature in generating the output. The models will also be tested through Data Augmentation to generate a greater number of examples for the underrepresented classes. Finally, Adversarial Models will be in- troduced to test the validity of the obtained explainability results and to verify their robustness.