Arguing Prism: An Argumentation Based Approach for Collaborative Classification in Distributed Environments

This paper focuses on collaborative classification, concerning how multiple classifiers learned from distributed data repositories, can come to reach a consensus. Interpretability, in this context, is favored for the reason that they can be used to identify the key features influencing the classification outcome. In order to address this problem, we present Arguing Prism, an argumentation based approach for collaborative classification. The proposed approach integrates the ideas from modular classification rules inductive learning and multi-agent dialogue games. In particular, argumentation is used to provide an interpretable classification paradigm in distributed environments, rather than voting mechanisms. The results of experiments reveal that Arguing Prism performs better than individual classifier agents voting schemes. Moreover, an interpretable classification can be achieved without losing much classification performance, when compared with ensemble classification paradigms. Further experiment results show that Arguing Prism out-performs comparable classification paradigms in presence of inconsistent data, due to the advantages offered by argumentation.