The Bacterial Foraging Optimisation Algorithm using Prototype Selection and Prototype Generation for Data Classification

A new emerging nature-inspired algorithm named Bacterial Foraging Optimisation Algorithm (BFOA) that mimics the foraging behaviour of E. coli bacteria has drawn lots of attention from other researchers due to its high convergence rate and global search capability compared to others. Technically, BFO...

Full description

Saved in:
Bibliographic Details
Main Author: Faizol, Bin Mohd Suria
Format: Thesis
Language:English
Published: 2020
Subjects:
Online Access:http://ir.unimas.my/id/eprint/30023/1/The%20Bacterial..ft.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:A new emerging nature-inspired algorithm named Bacterial Foraging Optimisation Algorithm (BFOA) that mimics the foraging behaviour of E. coli bacteria has drawn lots of attention from other researchers due to its high convergence rate and global search capability compared to others. Technically, BFOA has been applied as supplementary algorithm for optimizing weight, parameters for other classifier algorithms and selecting optimised features for other classifiers. However, none of the available works had proposed BFOA as a classification algorithm despite of its good performance. Thus, this study aims to adopt and modify the BFOA into Instance Selection (IS) classifier by manipulating its global search capability and high convergence rate for data classification problem. There are two modified instance-based classifiers of BFOA were developed in this study. Both classifiers are inspired based on Prototype Selection (PS) and Prototype Generation (PG) known as BFOA-S and BFOA-G respectively. BFOA-G statistically outperformed BFOA-S by achieving 80.73% in the average testing accuracy with 5.16% average storage requirement, and 3 times faster in term of time complexity against BFOA-S in the benchmark experiment using 42 datasets. In addition, BFOA-G also performed well against ten existing IS algorithms by obtaining 83.1% in the average accuracy with 95.51% reduction rate and ranked first in the comparison study. On the other hand, BFOA-S showed competitive performance in the comparison against ten existing IS algorithms by obtaining 96.25% in reduction rate and ranked third in the ranking. Therefore, we conclude that the proposed BFOA-G is the best algorithm in the study, and PG approach is recommended for further development of BFOA as an IS algorithm.