• affective computing
• counterfactual explanations
• eXplainable Artificial Intelligence
• latent space

This thesis presents a post-hoc XAI-based method for generating local counterfactual explanations for any multi-class classifiers for tabular datasets (model-agnostic). The proposed approach exploits lower-dimensional latent spaces to produce class-specific counterfactuals. To assess its effectiveness, the method is evaluated on both benchmark datasets and a real-world dataset (i.e., physiological data for emotion recognition). The proposed approach is evaluated using different measures such as counterfactual generation capability, acceptability, specificity, minimality, counterfactual anomaly score, and execution time.