Modified particle swarm optimization (MPSO) optimized CNN’s hyperparameters for classification

Murinto, Murinto and Sri, Winiarti (2025) Modified particle swarm optimization (MPSO) optimized CNN’s hyperparameters for classification. [Artikel Dosen]

Text Modified_particle_swarm_optimi.pdf - Published Version Download (785kB)

Abstract

This paper proposes a convolutional neural network architectural design approach using the modified particle swarm optimization (MPSO) algorithm. Adjusting hyper-parameters and searching for optimal network
architecture from convolutional neural networks (CNN) is an interesting challenge. Network performance and increasing the efficiency of learning models on certain problems depend on setting hyperparameter values, resulting in large and complex search spaces in their exploration. The use
of heuristic-based searches allows for this type of problem, where the main contribution in this research is to apply the MPSO algorithm to find the optimal parameters of CNN, including the number of convolution layers, the filters used in the convolution process, the number of convolution
filters and the batch size. The parameters obtained using MPSO are kept in the same condition in each convolution layer, and the objective function is evaluated by MPSO, which is given by classification rate. The optimized architecture is implemented in the Batik motif database. The research found that the proposed model produced the best results, with a classification rate higher than 94%, showing good results compared to other state-of-the-art approaches. This research demonstrates the performance of the MPSO algorithm in optimizing CNN architectures, highlighting its potential for improving image recognition tasks.

Item Type:	Artikel Dosen
Subjects:	T Technology > T Technology (General)
Divisi / Prodi:	Faculty of Industrial Technology (Fakultas Teknologi Industri) > S1-Informatics Engineering (S1-Teknik Informatika)
Depositing User:	murinto murinto
Date Deposited:	26 Aug 2025 06:14
Last Modified:	26 Aug 2025 06:14
URI:	http://eprints.uad.ac.id/id/eprint/86510

Actions (login required)

View Item