Multiview Fully Interpretable TSK Fuzzy Classifier Enhanced by Multiview Accompanying GMMs

While interpretable multiview classification is accomplished by recently developed fully interpretable Takagi–Sugeno–Kang (TSK) classifier fully interpretable and more generalization (FIMG)-TSK along each view, this study attempts to enhance multiview classification performance by leveraging the corresponding Gaussian mixture model (GMM) of FIMG-TSK along each view as an accompanying classifier along the accompanying view of each view. The basic idea behind this study is to have an individual multiview ensemble learning not only for FIMG-TSK along each view but also for the corresponding GMM along each accompanying view, and then take the averaging strategy on them both to form a novel multiview fully interpretable TSK fuzzy classifier called MV-FIMG-TSKs. By maximizing the concisely-expressed output variance of FIMG-TSK along each view and simultaneously keep consistent as much as possible between all the accompanying views, the learning objective of MV-FIMG-TSKs is designed to compromise all the views. Then, it is quickly optimized to have an analytical yet global solution to the weights of all rules along each view. Enhanced generalization capability of the proposed MV-FIMG-TSKs is theoretically assured and experimentally manifested by the experimental results on 14 binary datasets.