UTHM Institutional Repository

Component-wise analysis of metaheuristic algorithms for novel fuzzy-meta classifier

Talpur, Kashif Hussain (2018) Component-wise analysis of metaheuristic algorithms for novel fuzzy-meta classifier. Doctoral thesis, Universiti Tun Hussein Onn Malaysia.

[img]
Preview
Text
COMPONENT-WISE ANALYSIS OF METAHEURISTIC ALGORITHMS FOR NOVEL FUZZY-META CLASSIFIER.pdf

Download (6MB) | Preview

Abstract

Metaheuristic research has proposed promising results in science, business, and engineering problems. But, mostly high-level analysis is performed on metaheuristic performances. This leaves several critical questions unanswered due to black-box issue that does not reveal why certain metaheuristic algorithms performed better on some problems and not on others. To address the significant gap between theory and practice in metaheuristic research, this study proposed in-depth analysis approach using component-view of metaheuristic algorithms and diversity measurement for determining exploration and exploitation abilities. This research selected three commonly used swarm-based metaheuristic algorithms – Particle Swarm Optimization (PSO), Artificial Bee Colony (ABC), and Cuckoo Search (CS) – to perform component-wise analysis. As a result, the study able to address premature convergence problem in PSO, poor exploitation in ABC, and imbalanced exploration and exploitation issue in CS. The proposed improved PSO (iPSO), improved ABC (iABC), and improved CS (iCS) outperformed standard algorithms and variants from existing literature, as well as, Grey Wolf Optimization (GWO) and Animal Migration Optimization (AMO) on ten numerical optimization problems with varying modalities. The proposed iPSO, iABC, and iCS were then employed on proposed novel Fuzzy-Meta Classifier (FMC) which offered highly reduced model complexity and high accuracy as compared to Adaptive Neuro-Fuzzy Inference System (ANFIS). The proposed three-layer FMC produced efficient rules that generated nearly 100% accuracies on ten different classification datasets, with significantly reduced number of trainable parameters and number of nodes in the network architecture, as compared to ANFIS.

Item Type: Thesis (Doctoral)
Subjects: T Technology > T55.4-60.8 Industrial engineering. Management engineering
Divisions: Faculty of Computer Science and Information Technology > Department of Software Engineering
Depositing User: Sabarina Che Mat
Date Deposited: 29 Feb 2020 12:03
Last Modified: 29 Feb 2020 12:03
URI: http://eprints.uthm.edu.my/id/eprint/12146
Statistic Details: View Download Statistic

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year