OPTIMASI SIMULTAN SELEKSI FITUR DAN HIPERPARAMETER BERBASIS STABILITY-AWARE COST FUNCTION DAN PUMA OPTIMIZER UNTUK KLASIFIKASI MULTI-KELAS CITRA TORAKS

Baiq Titik Haerani, NIM.: 24206051010 (2026) OPTIMASI SIMULTAN SELEKSI FITUR DAN HIPERPARAMETER BERBASIS STABILITY-AWARE COST FUNCTION DAN PUMA OPTIMIZER UNTUK KLASIFIKASI MULTI-KELAS CITRA TORAKS. Masters thesis, UIN SUNAN KALIJAGA YOGYAKARTA.

Preview	Text (OPTIMASI SIMULTAN SELEKSI FITUR DAN HIPERPARAMETER BERBASIS STABILITY-AWARE COST FUNCTION DAN PUMA OPTIMIZER UNTUK KLASIFIKASI MULTI-KELAS CITRA TORAKS) 24206051010_BAB-I_IV-atau-V_DAFTAR-PUSTAKA.pdf - Published Version Download (7MB) | Preview
	Text (OPTIMASI SIMULTAN SELEKSI FITUR DAN HIPERPARAMETER BERBASIS STABILITY-AWARE COST FUNCTION DAN PUMA OPTIMIZER UNTUK KLASIFIKASI MULTI-KELAS CITRA TORAKS) 24206051010_BAB-II_sampai_SEBELUM-BAB-TERAKHIR.pdf - Published Version Restricted to Registered users only Download (17MB) | Request a copy

Abstract

Deep learning-based chest X-ray image classification generally produces high-dimensional feature representations, requiring feature selection and hyperparameter optimization to enhance the model's efficiency and performance. The wrapper-based simultaneous optimization approach has been widely used, but most still rely on an objective function that only considers classification error, thus not taking into account performance stability during the validation process. This research aims to develop and evaluate the Proposed Stability-Aware PUMA Framework, which is a simultaneous optimization framework that integrates deep feature selection and hyperparameter optimization using the Puma Optimizer with the Stability-Aware Cost Function (SACF) as the objective function. SACF combines classification error, error variance, and feature ratio to simultaneously optimize classification accuracy, validation performance consistency, and model complexity. Experiments were conducted on 532 four-class chest X-ray images. A total of 1,280 deep features were extracted using MobileNetV2 as a fixed feature extractor, and then simultaneously optimized along with the hyperparameters of Support Vector Machine (SVM) and Random Forest (RF). Evaluation includes testing classification performance, ablation study, convergence analysis, and robustness testing against Gaussian noise and speckle noise, as well as statistical validation. The research results show that the proposed framework is capable of reducing more than 50% of deep features while maintaining competitive classification performance. The ablation study proves that the combination of classification error, error variance, and feature ratio yields better optimization performance compared to using each component separately. Additionally, the proposed framework demonstrates a more consistent optimization process, more stable convergence, and competitive resilience against image degradation. These findings indicate that the integration of SACF in simultaneous optimization can produce more balanced solutions by considering accuracy, stability, and feature representation efficiency, thereby potentially enhancing generalization capabilities in multi-class chest X-ray image classification.

Share this knowledge with your friends :

Actions (login required)

View Item