Application of Soft Voting Ensemble Learning in the Classification of the Tinea Capitis Skin Disease using ResNet-50 and EfficientNet-B2

Nicholas, Sandy (2026) Application of Soft Voting Ensemble Learning in the Classification of the Tinea Capitis Skin Disease using ResNet-50 and EfficientNet-B2. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) 22081010237.-cover.pdf Download (711kB)
	Text (Bab 1) 22081010237.-bab1.pdf Download (161kB)
	Text (Bab 2) 22081010237.-bab2.pdf Restricted to Repository staff only until 6 April 2028. Download (609kB) \| Request a copy
	Text (Bab 3) 22081010237.-bab3.pdf Restricted to Repository staff only until 6 April 2028. Download (807kB) \| Request a copy
	Text (Bab 4) 22081010237.-bab4.pdf Restricted to Repository staff only until 6 April 2028. Download (814kB) \| Request a copy
	Text (Bab 5) 22081010237.-bab5.pdf Download (149kB)
	Text (Daftar Pustaka) 22081010237.-daftarpustaka.pdf Download (160kB)
	Text (Lampiran) 22081010237.-lampiran.pdf Restricted to Repository staff only until 6 April 2028. Download (961kB) \| Request a copy

Abstract

Tinea Capitis skin disease (scalp ringworm) is a dermatophyte fungal infection with a high prevalence of disease distribution, particularly among the pediatric age group, male gender, and rural communities, characterized by a rapid transmission risk. Diagnosing Tinea Capitis faces challenges involving high dermoscopy equipment costs and the subjectivity of medical image interpretation, exacerbated by visual similarities between inflammatory and non-inflammatory subtypes (Kerion, Pustular Diffuse, Black Dot, and Gray Patch). This study aims to address these issues by proposing a Tinea Capitis subtype classification system based on Deep Learning using the Soft Voting Ensemble Learning paradigm, which combines the predictive strengths of Convolutional Neural Network (CNN) architectures, specifically the ResNet-50 and EfficientNet-B2 models. ResNet-50 architecture was selected for deep textural image feature extraction capabilities via Residual Blocks that mitigate the vanishing gradient problem, while the EfficientNet-B2 architecture was chosen for parameter efficiency and Compound Scaling mechanism, offering a balance between depth, width, and resolution. To maximize the performance of Soft Voting Ensemble Learning, contribution weight combination optimization is applied by comparing the deterministic Grid Search method and the stochastic Random Search method across three dataset splitting scenarios. System testing results indicate that Scenario 1 (70% training : 20% validation : 10% testing) yielded the most robust performance with the highest testing accuracy reaching 89.47%, where the Grid Search optimization proved consistent compared to Random Search in finding the optimal ensemble weight configuration (w1 = 0.45 for ResNet-50 architecture model and w2 = 0.55 for EfficientNet-B2 architecture model), and successfully corrected classification errors in the Pustular Diffuse class. The best architectural model results are subsequently implemented into a web-based interface system utilizing Laravel 12 and Tailwind CSS frameworks to facilitate direct use by medical personnel in the field. Performance evaluation using Accuracy, Precision, Recall, F1-Score, Specificity, and False Positive Rate evaluation metrics concludes that the integration of the optimized Soft Voting Ensemble Learning applied within a medical information system is capable of providing an accurate, efficient, and accessible diagnostic solution for medical personnel.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Diyasa, I Gede Susrama Mas	NIDN0019067008	igsusrama.if@upnjatim.ac.id
Thesis advisor	Maulana, Hendra	NIDN1423128301	hendra.maulana.if@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA76 Computer software
Q Science > QA Mathematics > QA76.6 Computer Programming
Q Science > QA Mathematics > QA76.87 Neural computers

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Sandy Nicholas

Date Deposited:

07 Apr 2026 01:31

Last Modified:

07 Apr 2026 02:52

URI:

https://repository.upnjatim.ac.id/id/eprint/50460