Hyperparameter optimization of breast ultrasound image classification models using ant colony optimization based on texture features

• Published in: Indonesian Journal of Electronics, Electromedical Engineering, and Medical Informatics Vol. 7; no. 4; pp. 673 - 686
• Main Authors: Fauzi, Ahmad, Lubis, Lukmanda Evan, Wandira, Raju, Musthofa, Syarto
• Format: Journal Article
• Language: English
• Published: 23.10.2025
• ISSN: 2656-8624; 2656-8624
• DOI: 10.35882/ijeeemi.v7i4.266

Breast cancer is one of the most prevalent types of cancer in humans and has the highest cumulative risk compared to other types of cancer. Accurate diagnosis and efficient intervention for this disease are very important to improve patient survival. This study aims to optimize machine learning algorithms using a limited number of features in order to produce an efficient breast cancer classification model that remains competitive with deep learning–based models. Furthermore, this study is expected to assist pathologists and doctors in the treatment of breast cancer. The dataset used in this study consists of three classes (benign, malignant, and normal) with a total of 780 breast ultrasound images from 600 patients. All images were processed and augmented to enrich data variation before modeling using five classification algorithms: Random Forest, SVM, Decision Tree, Gradient Boosting, and k-NN. Modeling was conducted in two scenarios: without optimization and with hyperparameter optimization using the Ant Colony Optimization algorithm. The results showed that the GLCM angle orientation had a relatively small effect on model performance. The best accuracy for each orientation was achieved with k-NN+ ACO (0.95 at 00), SVM+ACO (0.94 at 450), SVM+ACO (0.90 at 900), and RF+ACO (0.95 at 1350).