Feature reduction techniques for software bug prediction.

Developing a software entails writing thousands of lines of code. For ensuring quality of the software, this code must be fault free (should perform as it is intended to do). Software faults result in wastage of effort and resources used for developing it. Software bug prediction is a process in the initial period of Software Development Life Cycle (SDLC) which predicts bug-prone modules in a software. Various Machine Learning (ML) methods and feature reduction techniques have been employed for better fault prediction. In this paper six feature reduction techniques have been employed on five software bug datasets of AEEEM software repository in association with random forest-based ensemble classifier. SMOTE and Stratified 10-fold cross validation are used to improve performance of bug prediction model. Three performance metrics (ROC-AUC, F1-Score and accuracy) are ex-tracted for evaluating different dimensions of prediction. Feature agglomeration and Sparse principal component analysis performed better on these datasets for feature reduction. [ABSTRACT FROM AUTHOR]