A Novel Wrapper-based Feature Selection Model using Fennec Fox Optimization for Software Fault Prediction

Software fault prediction helps software developers with detecting software defects ahead of time. Early detection of software defects saves the effort and money that would be needed to rectify and redesign the software modules. Machine learning algorithms (classifiers) are used to predict these defects using previous data of faults and software metrics as independent features. However, these data are very huge, high-dimensional, and have a lot of noise and unwanted features. So, the process of feature selection (FS) is required to select only the relevant features and obtain the most crucial information from the data. In this paper, a novel wrapper-based algorithm for FS called Feature Selection using Fennec Fox Optimization (FSFFO) is proposed. The FSFFO model was embedded with four well-known classifiers – Naïve Bayes, Decision Tree, KNN, and Quadratic Discriminant Analysis. The proposed FSFFO model’s performance was compared with previously developed feature selection models such as FSPSO, FSDE, FSGA, and FSACO. For experimentation, 12 open-source software defect datasets were used. The results from experiments showed that the proposed FSFFO model significantly outperformed all the other FS models taken into consideration in the majority of test cases. FSFFO gave higher accuracy of classification, and the number of selected features was less than 50% of the original number of features, showing its superiority.