An Optimized FL-XAI model for secured and trustworthy candidate selection

Abstract

Reliability and trustworthiness are the two pillars of decision support systems deployed in the selection process of automated candidates. The automation should ensure the selection's correctness and the decision's fairness. Conventional models work around fuzzy-based systems, exploiting multi-criteria decision support systems. Here, we propose a procedure combining the advantages of Federated Learning (FL) and Explainable Artificial Intelligence (XAI), ensuring privacy, reliability, and fairness in selecting candidates. We propose an architecture in which the exploitation of FL provides more accurate classification results while XAI provides a trustworthy and reliable representation of the candidate selection through decision plots. The SHAPELY model is used in the proposed work for explanation. Results and comparisons with several machine learning (ML) algorithms show the superiority of the proposed architecture. FL can reach an accuracy of 96%, thus confirming the validity of the proposed approach for providing an automated and well-explained candidate selection process.