Machine Learning Classifiers Explanations with Prototype Counterfactual

Machine learning applications have rapidly entered our day-to-day lives, and the stakes of input from these applications have increased for various highly valued decision-making processes. Explanation of these applications has been a prime focus in recent years for better interpretability, fairness, and reliability. In this paper, using the prototype, we presented a novel approach for counterfactual generation, resulting in better proximal instances with fewer variable characteristics and interpretability metrics that enable a better understanding of the models. We have empirically presented the result of the counterfactual generator and interpretability metric on tabular and image datasets, i.e., the Adult-Income, MNIST, and Breast Cancer datasets. We compared our proposed results to KD-Tree, and the Genetic Algorithm, demonstrating improved interpretability and well-approximated counterfactuals.