Bayu, Airlangga Bayu Taqwa (2026) Pengembangan Sistem Klasifikasi Lesi Kulit Berbasis Citra Dermoskopi Multimodal Menggunakan Convolutional Neural Network (CNN) Dan Generative Adversarial Network (GAN) Untuk Mengatasi Imbalanced Dataset. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5026221204-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5026221204-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (7MB) |
Abstract
Sebagian besar penelitian klasifikasi lesi kulit berbasis deep learning masih terbatas pada pendekatan single-modal yang hanya menggunakan citra dermoskopi, namun belum secara menyeluruh menangani ketidakseimbangan data pada kedua modalitas (citra dan data klinis). Untuk mengatasi kesenjangan tersebut, penelitian ini mengembangkan sistem klasifikasi delapan jenis lesi kulit dari dataset ISIC 2019 dengan pendekatan multimodal yang menggabungkan citra dan data klinis. Ketidakseimbangan data pada kedua modalitas ditangani dengan Generative Adversarial Network (GAN) untuk menambah variasi citra pada kelas minoritas dan SMOTE untuk menyeimbangkan data tabular. Citra diproses menggunakan base model CNN dan head model yang dirancang khusus, sedangkan data tabular diproses melalui normalisasi, encoding, dan dense layer sebelum digabungkan pada fusion layer. Model dievaluasi menggunakan confusion matrix untuk menilai peningkatan performa dibanding model lain. Model terbaik akan diintegrasikan ke dalam prototipe website berbasis Flask untuk mendukung deteksi dini lesi kulit yang mudah diakses masyarakat. Penelitian ini diharapkan dapat berkontribusi dalam pengembangan metode klasifikasi lesi kulit berbasis CNN multimodal dengan mengombinasikan GAN dan SMOTE untuk meningkatkan akurasi pada dataset yang tidak seimbang, serta mendukung sistem pendukung diagnosis yang lebih akurat dan objektif di Indonesia.
==================================================================================================================================
Most deep learning-based skin lesion classification studies are still limited to single-modal approaches that only use dermoscopy images, but have not comprehensively addressed the imbalanced data in both modalities (images and clinical data). To resolve the gap, this study developed a classification system for eight types of skin lesions from the ISIC 2019 dataset using a multimodal approach that combines images and clinical data. The imbalanced data in both modalities was addressed using a Generative Adversarial Network (GAN) to increase image variation in the minority class and SMOTE to balance the tabular data. Images were processed using a specially designed CNN base model and head model, while tabular data were processed through normalization, encoding, and dense layers before being combined in a fusion layer. The models were evaluated using a confusion matrix to assess performance improvements compared to other models. The best model will be integrated into a Flask-based website prototype to support easily accessible early detection of skin lesions. This research is expected to contribute to the development of multimodal CNN-based skin lesion classification methods by combining GAN and SMOTE to improve accuracy on imbalanced datasets, as well as supporting a more accurate and objective diagnostic support system in Indonesia.
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | Classification of Skin Lesions, Klasifikasi Lesi Kulit, Deep Learning, Convolutional Neural Network, Multimodal, Generative Adversial Network, SMOTE. |
| Subjects: | Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) |
| Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis |
| Depositing User: | Airlangga Bayu Taqwa |
| Date Deposited: | 05 Aug 2026 01:37 |
| Last Modified: | 05 Aug 2026 01:37 |
| URI: | http://repository.its.ac.id/id/eprint/141218 |
Actions (login required)
 |
View Item |