Human Skin Disease Classification Using Swin Transformer With the Development of a Knowledge-Based Treatment Recommendation System

Hartono, Anggara Falih (2026) Human Skin Disease Classification Using Swin Transformer With the Development of a Knowledge-Based Treatment Recommendation System. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) 22081010122-cover.pdf Download (545kB)
	Text (Bab 1) 22081010122-bab1.pdf Download (195kB)
	Text (Bab 2) 22081010122-bab2.pdf Restricted to Repository staff only until 25 June 2029. Download (461kB)
	Text (Bab 3) 22081010122-bab3.pdf Restricted to Repository staff only until 25 June 2029. Download (581kB)
	Text (Bab 4) 22081010122-bab4.pdf Restricted to Repository staff only until 25 June 2029. Download (725kB)
	Text (Bab 5) 22081010122-bab5.pdf Download (212kB)
	Text (Daftar Pustaka) 22081010122-daftarpustaka.pdf Download (182kB)
	Text (Lampiran) 22081010122-lampiran.pdf Restricted to Repository staff only until 25 June 2029. Download (139kB)

Abstract

Skin disease is one of the most common health problems and can significantly affect patients' quality of life. The diagnosis process is generally performed through visual observation of lesion characteristics, such as color, shape, and texture. However, visual similarities among different skin diseases often create challenges in the initial identification process. This study aims to develop a human skin disease classification system using the Swin Transformer architecture and a knowledge-based treatment recommendation system based on the Clinical Practice Guidelines published by the Indonesian Society of Dermatology and Venereology (PERDOSKI). The dataset consists of four skin disease classes, namely Acne, Melanoma, Tinea, and Verruca/Warts. The proposed model was developed using the Swin Transformer architecture with the best experimental configuration consisting of an 80:20 data split ratio, a learning rate of 1×10⁻⁴, a batch size of 32, and 100 training epochs. Transfer learning was applied to improve the model's capability in extracting discriminative visual features from skin lesion images. After the classification process, the system provides treatment recommendations according to the predicted disease category. Experimental results show that the proposed model achieved an accuracy of 92.27%, a weighted precision of 92.25%, a weighted recall of 92.27%, and a weighted F1-score of 92.25%. In addition, the model obtained a macro F1-score of 92.22%, indicating a balanced classification performance across all classes. The Melanoma class achieved the highest performance with an F1-score of 96.52%, while the Verruca/Warts class obtained an F1-score of 87.50%. These findings demonstrate that the Swin Transformer architecture is highly effective for multi-class skin disease classification and can be integrated with a recommendation system to support preliminary disease identification and provide treatment-related information for users.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Puspaningrum, Eva Yulia	NIP198907052021212002	evapuspaningrum.if@upnjatim.ac.id
Thesis advisor	Muttaqin, Faisal	NIP198512312021211009	faisalmuttaqin.if@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA75 Electronic computers. Computer science

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Anggara Falih Hartono

Date Deposited:

25 Jun 2026 06:11

Last Modified:

25 Jun 2026 06:59

URI:

https://repository.upnjatim.ac.id/id/eprint/54206