Optimizing Neural Network Training for Image Classification Using Hybrid PSO-GABackpropagation Algorithms
Essang Samuel Okon1, Stephen I. Okeke2, Ante Jackson Efiong3, Francis Runyi, Fadugba4, Sunday Emmanuel5, Augustine Ogbaji Otobi6
1Essang Samuel Okon, Department of Mathematics and Computer Science, Arthur Jarvis University, Akpabuyo, Nigeria.
2Stephen I. Okeke, International Institute for Machine Learning, Robotics and Artificial Intelligence, David Umahi Federal University of Health Sciences, Uburu, Ebonyi State, Nigeria, Department of Industrial Mathematics and Health Statistics, David Umahi Federal University of Health Sciences, Uburu, Ebonyi State, Nigeria.
3Ante Jackson Efiong, Department of Mathematics, TopFaith University, Mkpatak, Nigeria.
4Francis Runyi, Federal Polytechnic Ugep, Nigeria.
5Fadugba, Sunday Emmanuel, Department of Mathematics, Ekiti State University, Ado Ekiti, Nigeria.
6Augustine Ogbaji Otobi, University of Calabar Zoe-Sprout International Ado Ekiti, Nigeria.
Manuscript received on 08 January 2023 | Revised Manuscript received on 29 January 2023 | Manuscript Accepted on 15 March 2023 | Manuscript published on 30 March 2023 | PP: 15-26 | Volume-12 Issue-4, March 2023 | Retrieval Number: 100.1/ijitee.C105014030225 | DOI: 10.35940/ijitee.C1050.12040323
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Image classification has become a crucial application of artificial intelligence, utilising neural networks to analyse and understand intricate visual input. However, traditional training methods such as backpropagation face significant challenges, including susceptibility to local minima and reliance on optimal hyperparameters. This study investigates a hybrid optimisation system that integrates Particle Swarm Optimisation (PSO), Genetic Algorithms (GA), and backpropagation to overcome these restrictions. The hybrid PSO-GA-backpropagation algorithm combines the global search abilities of PSO and GA with the local optimisation of backpropagation, facilitating efficient and precise neural network training. The proposed method was applied to both simulated and real-world datasets, including a binary image classification problem with noisy geometric shapes, as well as standard datasets such as MNIST and CIFAR-10. The hybrid strategy demonstrated enhanced performance, with greater accuracy (92%) and greater resilience to noise compared to independent methods. Essential visualisations, such as loss curves and ROC assessments, underscore the algorithm’s proficient convergence and classification efficacy. Additionally, the analysis of noise sensitivity and parameter optimisation was conducted to guarantee adaptation to various conditions. This study enhances hybrid optimisation methods for neural network training, offering a scalable and resilient solution for high-dimensional image categorisation challenges.
Keywords: Hybrid Optimization, Particle Swarm Optimization (PSO), Genetic Algorithms (GA), Backpropagation, Image Classification, Neural Network Training
Scope of the Article: Artificial Intelligence & Methods