CLASSIFICATION OF MALARIA USING MICROSCOPIC IMAGES OF BLOOD CELLS WITH A HYBRID CNN-SWIN TRANSFORMER ARCHITECTURE

RAMADHANI, MUHAMMAD NABIL (2026) CLASSIFICATION OF MALARIA USING MICROSCOPIC IMAGES OF BLOOD CELLS WITH A HYBRID CNN-SWIN TRANSFORMER ARCHITECTURE. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) COVERR.pdf Download (447kB)
	Text (Bab 1) BAB 1.pdf Download (234kB)
	Text (Bab 2) BAB 2.pdf Restricted to Repository staff only until 20 July 2028. Download (679kB)
	Text (Bab 3) BAB 3.pdf Restricted to Repository staff only until 20 July 2028. Download (945kB)
	Text (Bab 4) BAB 4.pdf Restricted to Repository staff only until 20 July 2028. Download (2MB)
	Text (Bab 5) BAB 5.pdf Download (203kB)
	Text (Daftar Pustaka) DAFTAR PUSTAKA.pdf Download (163kB)

Abstract

Malaria remains a significant public health threat in endemic regions, including Indonesia, while manual microscopic diagnosis remains prone to subjectivity and is time-consuming. This study applies binary classification of malaria (infected vs. uninfected) on microscopic blood cell images using a Hybrid CNN–Swin Transformer architecture, which combines CNN's local feature extraction with Swin Transformer's global context modeling. The images underwent preprocessing consisting of resizing to 224×224 pixels, Z-Score normalization, and on-the-fly augmentation, with performance optimized through hyperparameter tuning using Grid Search. Evaluation results on the best configuration for each model show that the Swin Transformer achieved an accuracy of 96.57% and F1-score of 96.54% (70:15:15), the CNN achieved an accuracy of 96.01% and F1-score of 96.01% (70:15:15), while the Hybrid model achieved an accuracy of 96.06% and F1-score of 96.01% (70:15:15). The slightly lower performance of the Hybrid model is attributed to spatial information reduction at the bridge layer, the architecture's sensitivity to hyperparameter combinations, and the characteristics of the single-cell dataset, in which the Swin Transformer's global context alone is already sufficient for classification. Interpretability was analyzed using Grad-CAM, which showed the model's focus consistently on the morphology of the Plasmodium parasite as the biologically relevant region. The system was implemented as a web-based prototype using Streamlit to support fast and interactive malaria detection.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Al Haromainy, Muhammad Muharrom	0701069503	muhammad.muharrom.if@upnjatim.ac.id
Thesis advisor	Wahanani, Henni Indah	0022097811	henniendah.if@upnjatim.ac.id

Subjects:

T Technology > T Technology (General)

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Muhammad Nabil Ramadhani

Date Deposited:

22 Jul 2026 07:48

Last Modified:

22 Jul 2026 07:48

URI:

https://repository.upnjatim.ac.id/id/eprint/56571