Implementasi Efficientvit dengan Adaptive Fine-Tuning untuk Klasifikasi Penyakit Kulit Autoimun

Adnanto, Hafiyan Fazagi (2025) Implementasi Efficientvit dengan Adaptive Fine-Tuning untuk Klasifikasi Penyakit Kulit Autoimun. Undergraduate thesis, UPN "Veteran" Jawa Timur.

	Text (Cover) 21081010124-Cover-pages.pdf Download (1MB)
	Text (Bab 1) 21081010124-Bab 1.pdf Download (273kB)
	Text (Bab 2) 21081010124-Bab 2.pdf Restricted to Repository staff only until 5 December 2028. Download (1MB)
	Text (Bab 3) 21081010124-Bab 3.pdf Restricted to Repository staff only until 5 December 2028. Download (1MB)
	Text (Bab 4) 21081010124-Bab 4.pdf Restricted to Repository staff only until 5 December 2028. Download (2MB)
	Text (Bab 5) 21081010124-Bab 5.pdf Download (184kB)
	Text (Daftar Pustaka) 21081010124-Daftar pustaka.pdf Download (201kB)
	Text (Lampiran) 21081010124-Lampiran.pdf Download (601kB)

Abstract

This study aims to enhance the performance of autoimmune skin disease image classification by implementing the EfficientViT architecture combined with an adaptive fine-tuning strategy. EfficientViT integrates the computational efficiency of Convolutional Neural Networks (CNNs) with the global representation capability of Vision Transformers (ViTs), resulting in a lightweight yet highly expressive model. The adaptive fine-tuning strategy is applied through two main mechanisms: an adaptive unfreezing layer that progressively unfreezes model layers according to the number of training epochs, and a learning rate scheduler based on cosine annealing with warmup to dynamically adjust the learning rate during training. The dataset consists of five image classes such as Lichen, Lupus, Psoriasis, Vitiligo, and Normal Skin collected from the Kaggle platform. After undergoing preprocessing steps including resizing, augmentation, and normalization, the model was evaluated using a classification report comprising accuracy, precision, recall, and F1-score metrics. Experimental results show that EfficientViT with adaptive fine-tuning achieved the best accuracy of 99.33%, with weighted average precision, recall, and F1-score values of 0.9933 each, outperforming the baseline model without adaptive fine-tuning (98.80% accuracy). These findings demonstrate that integrating adaptive fine-tuning through adaptive unfreezing layers and a cosine annealing with warmup scheduler enhances the training stability, computational efficiency, and overall accuracy of the EfficientViT model for autoimmune skin disease classification.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Sari, Anggraini Puspita	NIDN0716088605	anggraini.puspita.if@upnjatim.ac.id
Thesis advisor	Junaidi, Achmad	NIDN0710117803	achmadjunaidi.if@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA76.6 Computer Programming
T Technology > T Technology (General)

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Mr. Hafiyan Fazagi Adnanto

Date Deposited:

05 Dec 2025 08:55

Last Modified:

05 Dec 2025 09:00

URI:

https://repository.upnjatim.ac.id/id/eprint/48081