Modifikasi Genetic Algorithm Dua Tahap untuk Seleksi Fitur dan Optimasi Bobot Ensemble Classifier pada Klasifikasi Citra Ultrasonografi Payudara

Arif, Faren Haseena (2026) Modifikasi Genetic Algorithm Dua Tahap untuk Seleksi Fitur dan Optimasi Bobot Ensemble Classifier pada Klasifikasi Citra Ultrasonografi Payudara. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221031-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Kanker payudara merupakan salah satu penyebab kematian utama pada perempuan, sehingga deteksi dini melalui skrining menjadi krusial. Citra ultrasonografi payudara (BUS) banyak digunakan karena non-invasif, namun interpretasinya masih menghadapi tantangan akibat variabilitas citra dan keterbatasan fitur model konvensional. Penelitian ini mengembangkan sistem klasifikasi tiga kelas (jinak, ganas, normal) yang lebih akurat dan efisien melalui integrasi fusi citra dan optimasi algoritma evolusioner.
Metode yang diusulkan berupa pipeline yang mengintegrasikan fusi wavelet Haar antara citra BUS dan ground truth mask lesi sebelum ekstraksi fitur menggunakan MobileNet V1, dilanjutkan modifikasi genetic algorithm (GA) dua tahap. Tahap pertama mengoptimasi seleksi fitur melalui modifikasi operator crossover, seleksi adaptif, dan inisialisasi opposition-based learning (OBL); tahap kedua mengoptimasi bobot ensemble melalui grid search integer dan GA bilangan riil. Evaluasi mencakup 88 konfigurasi, sebelas modifikasi GA, empat strategi pembobotan, dua kondisi data, menggunakan dataset BUSI (780 citra) dengan validasi eksternal pada BUS-UCLM (683 citra).
Pada kondisi fused, konfigurasi terbaik (inisialisasi OBL, bobot [1,1,2,1]) mencapai F1-score makro 0,9778 dan akurasi 0,9808, dengan reduksi fitur 54,9%, unggul 0,62 poin di atas baseline. Pada kondisi raw, bobot yang sama mencapai F1-score makro 0,8833, atau 2,81 poin di atas baseline. Validasi pada BUS-UCLM menunjukkan performa konsisten dan kompetitif (F1-score makro 0,9134–0,9267). Fusi wavelet memberikan kontribusi performa terbesar, sementara modifikasi GA berbasis OBL dan crossover-seleksi konsisten meningkatkan seleksi fitur maupun pembobotan ensemble.
==============================================================================================================================
Breast cancer is one of the leading causes of death among women, making early detection through screening critical. Breast ultrasound (BUS) imaging is widely used for its non-invasive nature, yet interpretation remains challenging due to image variability and the limited discriminative power of conventional features. This study develops a more accurate and efficient three-class classification system (benign, malignant, normal) through image fusion and evolutionary algorithm-based optimization.
The proposed method is a pipeline integrating Haar wavelet fusion between BUS images and lesion ground truth masks prior to feature extraction using MobileNet V1, followed by a two-stage modified genetic algorithm (GA). The first stage optimizes feature selection through modified crossover operators, adaptive selection, and opposition-based learning (OBL) initialization; the second optimizes ensemble weights via integer grid search and real-valued GA. The evaluation covered 88 configurations—eleven GA modifications, four weighting strategies, two data conditions—using the BUSI dataset (780 images) with external validation on BUS-UCLM (683 images).
Under the fused condition, the best configuration (OBL initialization, weights [1,1,2,1]) achieved a macro F1-score of 0.9778 and accuracy of 0.9808, with 54.9% feature reduction, 0.62 points above baseline. Under the raw condition, the same weighting achieved a macro F1-score of 0.8833, 2.81 points above baseline. External validation on BUS-UCLM showed consistent, competitive performance (macro F1-score 0.9134–0.9267). Wavelet fusion yielded the greatest performance gain, while OBL-based and crossover-selection GA modifications consistently improved both feature selection and ensemble weighting.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	klasifikasi citra ultrasonografi payudara, genetic algorithm, opposition-based learning, ensemble classifier, MobileNet V1, fusi wavelet, breast ultrasound image classification, genetic algorithm, opposition-based learning, ensemble classifier, MobileNet V1, wavelet fusion.
Subjects:	R Medicine > R Medicine (General) > R858 Deep Learning
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Faren Haseena Arif
Date Deposited:	24 Jul 2026 02:56
Last Modified:	24 Jul 2026 02:58
URI:	http://repository.its.ac.id/id/eprint/136984

Actions (login required)

View Item