Optimasi Hyperparameter Pada Metode Convolutional Neural Network Untuk Klasifikasi Jenis Penyakit Kanker Kulit Menggunakan Bayesian Optimization

Nuzula, Muhammad Iqbal Firdaus (2025) Optimasi Hyperparameter Pada Metode Convolutional Neural Network Untuk Klasifikasi Jenis Penyakit Kanker Kulit Menggunakan Bayesian Optimization. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6025221050-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Penelitian ini bertujuan untuk meningkatkan kinerja model Convolutional Neural Network (CNN) dalam klasifikasi jenis kanker kulit melalui optimasi hyperparameter menggunakan pendekatan Bayesian Optimization. Empat arsitektur CNN digunakan, yaitu EfficientNetV2S, EfficientNetV2M, EfficientNetV2L, dan ResNet50V2, masing-masing diuji dalam dua skenario pelatihan dengan augmentasi dan tanpa augmentasi. Evaluasi dilakukan menggunakan akurasi, presisi, recall, F1-score, dan Cohen’s Kappa. Hasil eksperimen menunjukkan bahwa Bayesian Optimization secara konsisten meningkatkan kinerja model pada hampir semua metrik. Model terbaik adalah EfficientNetV2S dengan hyperparameter tuning tanpa augmentasi, yang mencapai akurasi, presisi, recall, dan F1-score sebesar 0,98, serta Kappa 0,96. Model ResNet50V2 juga menunjukkan efisiensi tinggi dengan waktu pelatihan sekitar 2225 detik dan F1-score 0,96. Model EfficientNetV2M dan EfficientNetV2L mengalami peningkatan performa signifikan, dengan kenaikan akurasi lebih dari 10% sampai 17% dibandingkan parameter default. Secara keseluruhan, Bayesian Optimization terbukti efektif dalam menemukan kombinasi parameter optimal, sehingga meningkatkan akurasi, stabilitas, dan efisiensi pelatihan model CNN untuk klasifikasi citra medis.
========================================================================================================================================
This study aims to improve the performance of Convolutional Neural Network (CNN) models in classifying types of skin cancer through hyperparameter optimization using the Bayesian Optimization approach. Four CNN architectures were employed, namely EfficientNetV2S, EfficientNetV2M, EfficientNetV2L, and ResNet50V2, each evaluated under two training scenarios with augmentation and without augmentation. The models were assessed using accuracy, precision, recall, F1-score, and Cohen’s Kappa metrics. The experimental results indicate that Bayesian Optimization consistently enhances model performance across nearly all evaluation metrics. The best-performing model was EfficientNetV2S with hyperparameter tuning and without augmentation, achieving accuracy, precision, recall, and F1-score of 0.98, along with a Kappa score of 0.96. The ResNet50V2 model also demonstrated high efficiency, with a training time of approximately 2225 seconds and an F1-score of 0.96. EfficientNetV2M and EfficientNetV2L also showed significant performance improvements, with accuracy increases of more than 10% until 17% compared to their default parameter configurations. Overall, Bayesian Optimization has proven effective in identifying optimal hyperparameter combinations, thereby improving the accuracy, stability, and training efficiency of CNN models for medical image classification tasks.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Bayesian Optimization, CNN, Hyperparameter Tuning, Kanker kulit, Machine Learning, Bayesian Optimization, CNN, Hyperparameter Tuning, Skin Cancer, Machine Learning
Subjects:	Q Science T Technology > T Technology (General) > T58.5 Information technology. IT--Auditing T Technology > T Technology (General) > T58.62 Decision support systems T Technology > T Technology (General) > T59.7 Human-machine systems.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Muhammad Iqbal Firdaus Nuzula
Date Deposited:	31 Jul 2025 06:04
Last Modified:	31 Jul 2025 06:04
URI:	http://repository.its.ac.id/id/eprint/123962

Actions (login required)

View Item