Article Per Year (5 Year)
p-Index From 2021 - 2026
0.23
P-Index
This Author published in this journals
All Journal JTAM (Jurnal Teori dan Aplikasi Matematika)
Aris Thobirin
Department of Mathematics, Universitas Ahmad Dahlan
Mathematics

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

 1 
Computational Analysis of Xception and ConvMixer Architecture in Classification of Skin Disease Images using Geometric Transformation Maya Isafa Sam Saputri; Sugiyarto Surono; Aris Thobirin
JTAM (Jurnal Teori dan Aplikasi Matematika) Vol 10, No 3 (2026): July
Publisher : Universitas Muhammadiyah Mataram

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.31764/jtam.v10i3.37037

Abstract

This research seeks to evaluate and contrast the effectiveness of two deep learning models, Xception and ConvMixer, for classification of skin disease images. An experimental methodology was employed using the Massive Skin Disease. The data is divided into training, validation, and test data with a ratio of 80:10:10. The pre-processing stage includes resizing, normalization, and the application of geometric augmentation to improve visual variation in the training data. Both models were trained using equalized parameters so that comparisons were made objectively. The models were assessed through several evaluation metrics, including loss, accuracy, precision, recall, and F1-score metrics in a multi-class classification scheme. The results showed that Xception obtained a test accuracy of 99,70%, while ConvMixer achieved 94,60%. Additionally, Xception exhibits faster convergence and more stable inter-class performance consistency, while ConvMixer excels in compute time efficiency. This study contributes in the form of a comparative analysis of two modern architectures with training parameters that are equalized in the classification of skin diseases. However, the study is still limited to the use of a partial class and a single dataset, so further testing is needed to ensure the generalization capabilities of the model over a wider range of scenarios.
Co-Authors Maya Isafa Sam Saputri Sugiyarto Surono