Klasifikasi Citra Trachoma Menggunakan Seleksi Fitur SSD-AβHC dengan Mutual Information, Chaotic Maps, dan Fitness G-Mean

Robbani, Nisrina Salma (2026) Klasifikasi Citra Trachoma Menggunakan Seleksi Fitur SSD-AβHC dengan Mutual Information, Chaotic Maps, dan Fitness G-Mean. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221290-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Trachoma merupakan penyakit mata menular yang dapat menyebabkan gangguan penglihatan apabila tidak terdeteksi sejak dini. Klasifikasi citra medis masih menghadapi beberapa tantangan, seperti kemiripan visual antarkelas, distribusi data yang tidak seimbang, serta banyaknya fitur hasil ekstraksi deep learning yang tidak semuanya relevan. Kondisi tersebut dapat meningkatkan biaya komputasi dan menurunkan efektivitas proses seleksi fitur. Penelitian ini menggunakan L-CBAM-VGG16 sebagai model ekstraksi fitur dan SSD-AβHC sebagai metode seleksi fitur pada klasifikasi citra trachoma. Untuk memperkecil ruang pencarian, Mutual Information digunakan sebagai seleksi fitur berbasis filter sebelum proses seleksi fitur berbasis wrapper. Selain itu, penelitian ini juga menganalisis pengaruh inisialisasi populasi berbasis chaotic maps dan fungsi fitness berbasis G-Mean terhadap performa SSD-AβHC. Hasil pengujian menunjukkan bahwa konfigurasi terbaik diperoleh dari kombinasi L-CBAM-VGG16, MI-70, dan SSD-AβHC. Konfigurasi tersebut menghasilkan accuracy sebesar 0,9267, precision sebesar 0,9534, recall sebesar 0,7850, dan F1-score sebesar 0,8404 dengan 36 fitur terpilih dari 512 fitur awal. Hasil ini menunjukkan bahwa seleksi fitur Mutual Information mampu membantu SSD-AβHC menghasilkan subset fitur yang lebih ringkas, meningkatkan performa klasifikasi, dan mengurangi waktu komputasi. Namun, nilai recall dan G-Mean menunjukkan bahwa performa antarkelas masih belum sepenuhnya seimbang, terutama pada kelas minoritas.
=====================================================================================================================================
Trachoma is an infectious eye disease that can lead to visual impairment if not detected at an early stage. Medical image classification still faces several challenges, including high visual similarity between classes, imbalanced data distribution, and the high dimensionality of deep learning feature representations, many of which are irrelevant. These issues increase computational cost and reduce the effectiveness of the feature selection process.
This study employs L-CBAM-VGG16 as the feature extraction model and SSD-AβHC as the feature selection algorithm for trachoma image classification. To reduce the search space, Mutual Information is applied as a filter-based feature selection method prior to the wrapper-based feature selection process. In addition, this study investigates the effects of chaotic map-based population initialization and a G-Mean-based fitness function on the performance of SSD-AβHC. The experimental results show that the best performance is achieved by the combination of L-CBAM-VGG16, MI-70, and SSD-AβHC. This configuration achieves an accuracy of 0.9267, a precision of 0.9534, a recall of 0.7850, and an F1-score of 0.8404, while selecting only 36 features from the original 512 extracted features. The results demonstrate that Mutual Information effectively assists SSD-AβHC in generating a more compact feature subset, improving classification performance, and reducing computational time. However, the recall and G-Mean values indicate that the classification performance across classes remains imbalanced, particularly for the minority class.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	L-CBAM-VGG16, Mutual Information, Chaotic Maps, SSD-AβHC, Seleksi Fitur ========================================================== L-CBAM-VGG16, Mutual Information, Chaotic Maps, SSD-AβHC, Feature Selection
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Information Technology > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Nisrina Salma Robbani
Date Deposited:	24 Jul 2026 02:04
Last Modified:	24 Jul 2026 02:04
URI:	http://repository.its.ac.id/id/eprint/137211

Actions (login required)

View Item