Penerapan Convolutional Neural Network untuk Klasifikasi Cacar Monyet Berbasis Citra Menggunakan YOLOv11

Aripa, Salsabila Fatma (2025) Penerapan Convolutional Neural Network untuk Klasifikasi Cacar Monyet Berbasis Citra Menggunakan YOLOv11. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025211057-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Cacar monyet adalah penyakit zoonosis yang disebabkan oleh virus Monkeypox (MPXV) dari keluarga Poxviridae dan genus Orthopoxvirus. Meskipun tingkat kematian akibat monkeypox tergolong rendah, yakni sekitar 1% hingga 10% penyakit ini tetap menjadi ancaman kesehatan global. Diagnosis penyakit ini umumnya dilakukan melalui Polymerase Chain Reaction (PCR) yang dikenal karena sensitivitas dan akurasinya. Namun, keterbatasan ketersediaan uji PCR menghambat deteksi dini pada banyak wilayah. Sebagai alternatif, deteksi cacar monyet juga dapat dilakukan melalui klasifikasi citra medis berbasis AI. Penelitian ini mengimplementasikan algoritma Convolutional Neural Network (CNN) menggunakan model YOLOv11 versi klasifikasi (YOLOv11-cls) dengan varian nano (n) dan small (s) untuk mengklasifikasikan citra lesi kulit cacar monyet. Tiga dataset benchmark digunakan yaitu, MSID, MSLD, dan MCSI, dengan pembagian data melalui stratified sampling agar distribusi kelas tetap seimbang. Augmentasi citra dilakukan menggunakan pustaka Albumentations, sedangkan penyesuaian parameter dilakukan melalui Hyperparameter Tuning dengan pendekatan grid search menggunakan nested loop. Penelitian ini juga membandingkan performa YOLOv11 dengan arsitektur CNN lain seperti ResNet18 dan MobileNetV2. YOLOv11-cls varian small terpilih sebagai model terbaik dengan konfigurasi parameter image size 640×640, epochs 100, batch size 16, learning rate 0,001, dan optimizer SGD. Model terbaik kemudian divalidasi menggunakan metode K-Fold Cross Validation (k = 5) untuk memastikan stabilitas dan kemampuan generalisasi pada data yang berbeda. Evaluasi performa dilakukan dengan metrik accuracy, precision, recall, F1-score, confusion matrix, serta menghitung computational time. Hasil evaluasi menunjukkan model YOLOv11-cls versi small pada dataset MCSI mampu mencapai accuracy sebesar 0,9750, precision sebesar 0,9773, recall sebsar 0,9750, dan F1-score 0,9749 serta waktu komputasi selama 4,86 menit. Disusul dengan YOLOv11-cls versi medium dengan nilai accuracy sebesar 0,9500, precision 0,9545, recall 0,9500, F1-score 0,9484, dan waktu komputasi sekitar 4,80 menit. Hasil evaluas ini didapatkan bahwa YOLOV11-s memberikan performa yang palling tinggi dibandingkan model lainnya.
==================================================================================================================================
Monkeypox is a zoonotic disease caused by the Monkeypox virus (MPXV), which belongs to the Poxviridae family and the Orthopoxvirus genus. Although the fatality rate of monkeypox is relatively low, ranging from 1% to 10%, it remains a significant global health threat. Diagnosis is commonly performed through Polymerase Chain Reaction (PCR) testing, which is known for its high sensitivity and accuracy. However, limited access to PCR testing hampers early detection efforts in many regions. As an alternative, monkeypox detection can be conducted through medical image classification powered by artificial intelligence. This study implements a Convolutional Neural Network (CNN) algorithm using the YOLOv11 classification version (YOLOv11-cls) with nano (n) and small (s) variants to classify images of monkeypox skin lesions. Three benchmark datasets were utilized, namely MSID, MSLD, and MCSI, with data split using stratified sampling to maintain a balanced class distribution. Image augmentation was carried out using the Albumentations library, while Hyperparameter Tuning was conducted through a manual grid search approach using nested loops. This study also compares the performance of YOLOv11-cls with other CNN architectures such as ResNet18 and MobileNetV2. The YOLOv11-cls small variant was selected as the best-performing model with the following parameter configuration: image size of 640×640, 100 epochs, a batch size of 16, a learning rate of 0.001, and the SGD optimizer. The selected model was then validated using the K-Fold Cross Validation method (k = 5) to ensure stability and generalization capability across different data splits. Performance evaluation was conducted using metrics such as accuracy, precision, recall, F1-score, confusion matrix, and computational time measurement. The evaluation results show that the YOLOv11-cls small variant in the MCSI dataset achieved an accuracy of 0.9750, a precision of 0.9773, a recall of 0.9750, and an F1-score of 0.9749 with an average computational time of approximately 4.86 minutes. This was followed by the YOLOv11-cls medium variant with an accuracy of 0.9500, a precision of 0.9545, a recall of 0.9500, an F1-score of 0.9484, and an average computational time of around 4.80 minutes. These findings indicate that the YOLOv11-cls small variant delivers the highest performance compared to the other evaluated models.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	CNN, Cacar Monyet, Klasifikasi Gambar; YOLOv11, CNN, Mpox classification, Image Classification, YOLOv11
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing T Technology > T Technology (General) > T57.84 Heuristic algorithms.
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Mechanical Engineering > 21001-(S3) PhD Thesis
Depositing User:	Salsabila Fatma Aripa
Date Deposited:	25 Jul 2025 03:22
Last Modified:	25 Jul 2025 03:22
URI:	http://repository.its.ac.id/id/eprint/121743

Actions (login required)

View Item