Title: Fitness inheritance in multi-objective genetic algorithms: a case study on fuzzy classification rule mining

Authors: Harihar Kalia; Satchidananda Dehuri; Ashish Ghosh

Addresses: Department of Computer Science, Seemanta Engineering College, Mayurbhanj, 757086, Odisha, India ' Department of Information and Communication Technology, Fakir Mohan University, Balasore, 756019, Odisha, India ' Machine Intelligence Unit and Center for Soft Computing Research, Indian Statistical Institute, 203 BT Road, 700108, Kolkata, India

Abstract: The trade-off between accuracy and interpretability in fuzzy rule-based classifier has been examined in this paper by incorporating fitness inheritance in multi-objective genetic algorithms. The aim of this mechanism is to reduce the number of fitness evaluation by estimating the fitness value of the offspring individual from the fitness value of their parents. The multi-objective genetic algorithms with efficiency enhancement technique are a hybrid version of Michigan and Pittsburgh approaches. Fuzzy rules are represented by its antecedent fuzzy sets as an integer string of fixed length and a concatenated integer string of variable length. Our approach simultaneously maximises the accuracy of rule sets and minimises their complexity (i.e., maximisation of interpretability). As a result of adopting fitness inheritance, it minimises the overall time to generate rule set. The experimental outcome confirms that the proposed method reduces the computational cost, without compromising the quality of the results in a significant way.

Keywords: classification; fuzzy classification; multi-objective genetic algorithm; fitness inheritance; accuracy; interpretability.

DOI: 10.1504/IJAIP.2022.125235

International Journal of Advanced Intelligence Paradigms, 2022 Vol.23 No.1/2, pp.89 - 112

Received: 14 Feb 2017
Accepted: 12 Jun 2017
Published online: 05 Sep 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article