An Enhanced Evolutionary Software Defect Prediction Method Using Island Moth Flame Optimization

Ruba Abu Khurma,Ibrahim Aljarah,Hamad Alsawalqah,Robertas Damaševičius,Mohamed Abd Elaziz

doi:10.3390/math9151722

Abstract

Software defect prediction (SDP) is crucial in the early stages of defect-free software development before testing operations take place. Effective SDP can help test managers locate defects and defect-prone software modules. This facilitates the allocation of limited software quality assurance resources optimally and economically. Feature selection (FS) is a complicated problem with a polynomial time complexity. For a dataset with N features, the complete search space has 2N feature subsets, which means that the algorithm needs an exponential running time to traverse all these feature subsets. Swarm intelligence algorithms have shown impressive performance in mitigating the FS problem and reducing the running time. The moth flame optimization (MFO) algorithm is a well-known swarm intelligence algorithm that has been used widely and proven its capability in solving various optimization problems. An efficient binary variant of MFO (BMFO) is proposed in this paper by using the island BMFO (IsBMFO) model. IsBMFO divides the solutions in the population into a set of sub-populations named islands. Each island is treated independently using a variant of BMFO. To increase the diversification capability of the algorithm, a migration step is performed after a specific number of iterations to exchange the solutions between islands. Twenty-one public software datasets are used for evaluating the proposed method. The results of the experiments show that FS using IsBMFO improves the classification results. IsBMFO followed by support vector machine (SVM) classification is the best model for the SDP problem over other compared models, with an average G-mean of 78%.

Highlights

The software industry has recently undergone further development in various aspects related to the software development life-cycle (SDLC)
Software defect describes the error status that occurs at the program or system level which leads to erroneous results and unexpected actions and allows the system to behave in an unintended way [1]
This paper proposes the island model to enhance the binary variant of MFO (BMFO) for solving the Feature selection (FS) problem in the domain of software defect prediction

Summary

Introduction

The software industry has recently undergone further development in various aspects related to the software development life-cycle (SDLC). There are several reasons behind software defects [2] such as incomplete or ambiguous requirements due to miscommunication and misinterpretation during requirements elicitation, errors in assumptions and preliminary specifications, lack of knowledge in the domain, developers with insufficient practical experience and technical skills, poor programming logic, and so forth. This is a type of classification algorithm that belongs to a larger category of pattern recognition algorithms known as instance-based or lazy learning algorithms. Instead of conducting the generalization in an explicit training phase, they rely on computing the distance (similarities) between the unlabeled new query instance and its nearest k neighbors from the labeled training instances stored in memory. Assigning labels depends on the majority of votes obtained from the k closest neighbors for the required example. The comparison and the calculation of the closeness between points are done based on a predefined distance metric such as the Euclidean distance

Methods

Results

Conclusion