Hybrid Classifier for Software Defect Prediction by Using Filter-Based Feature Selection Techniques

Software’s capacity for error-free operation determines how reliable it is. Unfortunately, mistakes can happen at any stage of the software development process. Predicting software flaws early on in the development process is now of utmost importance in the field of software engineering. Using classification techniques, which are a popular strategy for predicting software faults, modules that are illustrated by a group of metrics or code properties are divided into defective or non-defective categories. However, there are low-quality, unreliable, redundant, and noisy data sources that have an adverse impact on the process of observing information and helpful patterns. Researchers must therefore use feature selection techniques to extract pertinent info from massive databases. In feature selection, the redundant and unnecessary features are eliminated in order to focus on the most significant ones. In this paper, we looked in to the impact of filter-based feature selection on classification techniques in software defect prediction. NASA and the Metric Data Program software repository both include publicly available datasets that were used in this study. Principal component analysis (PCA) and CFS was used to assess the dataset’s top discriminator features. The datasets were classified by using hybrid ensemble learning, for hybrid ensemble learning classifiers like DT, SVM, K-NN, NB, and LR were used. The experimental findings showed that it is preferable and should be encouraged to apply feature selection to datasets prior to classification in order to anticipate software defects.