A hierarchical prototype-based approach for classification

Abstract

In this paper, a novel hierarchical prototype-based approach for classification is proposed. This approach is able to perceive the data space and derive the multimodal distributions from streaming data at different levels of granularity in an online manner, based on which it further identifies meaningful prototypes to self-organize and self-evolve its hierarchical structure for classification. Thanks to the prototypebased nature, the system structure of the proposed classifier is highly transparent, and its learning process is of "one pass" type and computationally lean. Its decision-making process follows the "nearest prototype" principle and is fully explainable. The proposed approach is capable of presenting the learned knowledge from data in an easy-tointerpret prototype-based hierarchical form to users, and is an attractive tool for solving large-scale, complex real-world problems. Numerical examples based on various benchmark problems justify the validity and effectiveness of the proposed concept and general principles.

A self-training hierarchical prototype-based approach for semi-supervised classification

“…In this section, the general architecture, supervised learning and decision-making processes of the HP classifier [9] are briefly recalled to make this paper self-contained.…”

“…This subsection briefly recalls the supervised learning procedure of the HP classifier [9]. Since the prototypebased hierarchies are identified from labelled samples of each class separately, the learning process of the ith hierarchy is presented to avoid repetition ( = 1,2, … , ).…”

“…. Following the mode of [9], the radii of area of influence, ℎ around prototypes at different layers of the hierarchies are determined by the following expression (ℎ = 1,2, … , ):…”

“…One may specify the radius value setting based on preferences and specific requirements of the problems with the following constraint: 1 > 2 > ⋯ > . The supervised learning process is summarized as follows [9].…”

Highly interpretable hierarchical deep rule-based classifier

“…However, the majority of existing hierarchical classifiers suffer from the problem of inter-level error propagation due to the iterative structural optimization process involved during system identification [30,31]. More recently, a novel hierarchical prototype-based (HP) approach described in [32] presented an alternative tree structure for classification bypassing the aforementioned problem by self-organizing prototype-based hierarchies derived from training data per category individually. Nonetheless, the main issue with the HP classifier is that its systemmodel structuredepth and classification performancestructure are controlledinfluenced by externally controlled parameters.…”