Research Cell: An International Journal of Engineering Sciences

Harsimran Kaur ¹, Hardeep Singh ², Amitpal Singh Sohal ³

Affiliations
1 Research Scholar, Department of Computer Science, GNDU, Amritsar, IN
2 Professor, Department of Computer Science, GNDU, Amritsar, IN
3 Assitant Professor, Department of Computer Engineering and Technology, GNDU RC, Gurdaspur, IN

Abstract

Software fault prediction assists in identifying flaws in the early stages of software development and makes software testing more convenient and reliable. This study investigates the effect of the Principle Component Analysis techniques on software fault prediction models. It empirically compares the performance of six machine learning classifiers: Naïve Bayes, Random Forest, Logistic Regression, Support Vector Machine, Gradient Boosting, and Decision Tree with and without Principle Component Analysis (PCA). Furthermore, this paper aims to measure the capability of the software fault predictability in terms of accuracy, precision, f1 score, and AUC. The Area under Curve (AUC) Receiver Operating Characteristic (ROC) is used to check the validity of the models. The CAMEL dataset is used, which contains twenty-one Object-Oriented Software metrics available on the PROMISE repository. The Comparative evaluation indicates that all classifiers performed well with the Principal Component Analysis technique, whereas Random Forest and Decision Tree outperformed other classifiers.

Keywords

Machine Learning, Fault Prediction, Principle Component Analysis, AUC.

Full Text