Prolog-based agnostic explanation module for structured pattern classification

This paper presents a Prolog-based reasoning module to generate counterfactual explanations given the predictions computed by black-box classifier. Our approach comprises four well-defined stages that can be applied any structured pattern classification problem. Firstly, we pre-process dataset imputing missing values and normalizing numerical features. Secondly, transform features into symbolic ones using fuzzy clustering such extracted clusters are mapped an ordered set of predefined symbols. Thirdly, encode instances as Prolog rule nominal values, symbols, decision classes, confidence values. Fourthly, compute overall each fuzzy-rough theory handle uncertainty caused transforming quantities step comes with additional theoretical contribution new similarity function compare previously defined rules involving Finally, implement chatbot proxy between humans resolve natural language queries explanations. During simulations synthetic datasets, study performance our system when different operators functions.