Sine-Cosine Algorithm for Software Fault Prediction

Abstract

For developing an efficient and quality Software Fault Prediction (SFP) model, redundant and irrelevant features need to be removed. This task can be achieved, to a significant extent, with Feature Selection (FS) methods. Many empirical studies have been proposed on FS methods (Filter and Wrapper-based) and have shown effective results in reducing the problem of high dimensionality in metrics-based SFP models. This study evaluates the performance of novel wrapper-based Sine Cosine Algorithm (SCA) on five datasets of the AEEEM repository and compares the results with two metaheuristic techniques Genetic Algorithm (GA) and Cuckoo Search algorithm (CSA) on four different Machine Learning (ML) classifiers - Random Forest (RF), Support Vector Machine (SVM), Naïve Bayes (NB), and K-Nearest Neighbor (KNN). We found that the application of FS methods (SCA, GA & CSA) has improved the classifier performance. SCA has proved to be more efficient than GA and CSA in terms of lesser convergence time with the smallest subset of selected features and equivalent performance.