Computational argumentation based multi-agent consensual approach for semantically interpretable credit rating

It is well known that ensemble learning can make use of various classifiers, to capture various kinds of multimodal data features for credit rating in big data-driven financial risk analysis. However, it is very hard to provide semantic explanations for the evaluation outputs of intelligent systems when multiple classifiers come to a consensus. In this paper, a computational argumentation based multi-agent consensual approach has been proposed for credit rating. The method provides a natural means for justifying a decision, much like how people draw conclusions through argumentation. In particular, semantically interpretable consensus about the credit level of entities is provided with easily assimilated reasons in multiple classifier systems. The experimental evaluation over six benchmark real-world credit rating datasets shows that, compared with traditional strategies, not only better credit rating performance but also semantic explanation could be gained with the help of computational argumentation in ensemble learning for credit rating. Regarding multi-agent consensual credit rating, the competitive advantage offered by computational argumentation is that reasonable interpretation with arguments can be provided for financial decision makers, just like cognitive process in human intelligence.