An Ensemble Semi-Supervised Adaptive Resonance Theory Model With Explanation Capability for Pattern Classification

Most semi-supervised learning (SSL) models entail complex structures and iterative training processes as well face difficulties in interpreting their predictions to users. To address these issues, this paper proposes a new interpretable SSL model using the supervised unsupervised Adaptive Resonance Theory (ART) family of networks, which is denoted SSL-ART. Firstly, SSL-ART adopts an fuzzy ART network create number prototype nodes unlabeled samples. Then, it leverages ARTMAP structure map established target classes labeled Specifically, one-to-many (OtM) mapping scheme devised associate node with more than one class label. The main advantages include capability of: (i) performing online learning, (ii) reducing redundant through OtM minimizing effects noisy samples, (iii) providing explanation facility for users interpret predicted outcomes. In addition, weighted voting strategy introduced form ensemble model, WESSL-ART. Every member, i.e., SSL-ART, assigns {\color{black}a different weight} each based on its performance pertaining corresponding class. aim mitigate data sequences all members improve overall experimental results eighteen benchmark sets, three artificially generated real-world case study indicate benefits proposed WESSL-ART tackling pattern classification problems.