Studi Perbandingan Linear Discriminant Analysis (LDA) dan Singular Value Decomposition (SVD) dalam Mendeteksi Cacat Perangkat Lunak

Putri, Furstin Aprilavia (2026) Studi Perbandingan Linear Discriminant Analysis (LDA) dan Singular Value Decomposition (SVD) dalam Mendeteksi Cacat Perangkat Lunak. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221234-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Cacat perangkat lunak dapat menurunkan kualitas dan keandalan sistem serta menimbulkan kerugian besar, sehingga deteksinya sejak dini menjadi kebutuhan penting dalam rekayasa perangkat lunak. Prediksi cacat dengan pembelajaran mesin banyak digunakan untuk hal tersebut, tetapi dataset metrik seperti NASA Metrics Data Program (MDP) memiliki banyak fitur yang saling berkorelasi serta distribusi kelas yang sangat timpang sehingga menurunkan performa klasifikasi. Penelitian ini membandingkan dua teknik reduksi dimensi, yaitu Singular Value Decomposition (SVD) dan Linear Discriminant Analysis (LDA), untuk memprediksi modul cacat pada dua belas dataset NASA MDP, dengan seleksi fitur berbasis domain knowledge sebagai pembanding dan Conditional Tabular GAN (CTAB-GAN) untuk menyeimbangkan kelas. Pengujian disusun dalam lima skenario dan dievaluasi menggunakan repeated holdout dengan F1-Score macro sebagai acuan; skenario terakhir menguji konfigurasi terbaik tiap dataset menggunakan sembilan algoritma klasifikasi disertai penyetelan hyperparameter. CTAB-GAN menaikkan rata-rata F1-Score macro dari 0,5691 menjadi 0,5837 melalui peningkatan recall. SVD memperoleh rata-rata 0,6078, lebih tinggi daripada penggunaan seluruh fitur (0,5837), sedangkan LDA hanya 0,5657. Seleksi fitur mencapai rata-rata tertinggi 0,6233, dan kombinasi terbaik pada skenario terakhir mencapai 0,6451 dengan capaian puncak 0,7588 pada dataset KC3. Dengan demikian, SVD menunjukkan performa dan stabilitas yang lebih baik dibandingkan LDA, dan model terbaik diimplementasikan dalam antarmuka web berbasis Streamlit.
================================================================================================================================
Software defects can reduce the quality and reliability of a system and cause significant losses, so detecting them early is an important need in software engineering. Machine learning–based defect prediction is widely used for this purpose, but metric datasets such as the NASA Metrics Data Program (MDP) contain many correlated features and a highly imbalanced class distribution that degrade classification performance. This study compares two dimensionality reduction techniques, Singular Value Decomposition (SVD) and Linear Discriminant Analysis (LDA), for predicting defective modules across twelve NASA MDP datasets, with domain-knowledge-based feature selection as a comparator and a Conditional Tabular GAN (CTAB-GAN) to balance the classes. The experiments are arranged into five scenarios and evaluated using repeated holdout with the macro F1-Score as the reference; the final scenario tests each dataset's best configuration with nine classification algorithms and hyperparameter tuning. CTAB-GAN raised the average macro F1-Score from 0.5691 to 0.5837 through higher recall. SVD achieved an average of 0.6078, higher than using all features (0.5837), while LDA reached only 0.5657. Feature selection attained the highest average of 0.6233, and the best combination in the final scenario reached 0.6451, peaking at 0.7588 on the KC3 dataset. Thus, SVD demonstrated better performance and stability than LDA, and the best model was implemented in a Streamlit-based web interface.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	prediksi cacat perangkat lunak, reduksi dimensi, Singular Value Decomposition, Linear Discriminant Analysis, NASA Metrics Data Program, software defect prediction, dimensionality reduction.
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Furstin Aprilavia Putri
Date Deposited:	24 Jul 2026 03:52
Last Modified:	24 Jul 2026 03:52
URI:	http://repository.its.ac.id/id/eprint/136836

Actions (login required)

View Item