SOFTWARE DEFECT PREDICTION: EFFECT OF FEATURE SELECTION AND ENSEMBLE METHODS
No Thumbnail Available
Date
2018-09-10
Journal Title
Journal ISSN
Volume Title
Publisher
Federal University Wukari, Taraba State, Nigeria.
Abstract
Software defect prediction is the process of locating defective modules in software. It facilitates testing efficiency and consequently software quality. It enables a timely identification of fault-prone modules. The use of single classifiers and ensembles for predicting defects in software has been met with inconsistent results. Previous analysis say ensemble are often more accurate and are less affected by noise in datasets, also achieving lower average error rates than any of the constituent classifiers. However, inconsistencies exist in these various experiments and the performance of learning algorithms may vary using different performance measures and under different circumstances. Therefore, more research is needed to evaluate the performance of ensemble algorithms in software defect prediction. Adding feature selection reduces data sets with fewer features and improves the classifiers and ensemble performance over the datasets. The goal of this paper is to assess the efficiency of ensemble methods in software defect prediction using feature selection. This study compares the performance of four ensemble algorithms using 11 different performance metrics over 11 software defect datasets from the NASA MDP repository. The results indicate that feature selection and use of ensemble methods can improve the classification results of software defect prediction. Bagged ensemble models have the best results. In addition, Voting and Stacking also performed better than individual base classifiers. In terms of single classifier, SMO performs best as it outperformed Decision Tree (J48), MLP, and KNN with and without feature selection. Thus, it can be derived that feature selection can help improve the accuracy of both individual classifiers and ensemble methods by removing noisy and inconsistent features in the datasets.
Description
Keywords
Machine Learning, Data Mining, Software Defects Prediction, Software Development Process, Software Quality Assurance, Software Engineering
Citation
Mabayoje, M. A., Balogun, A. O., Bajeh, A. O., & Musa, B. A. (2018): Software Defect Prediction: Effect of Feature Selection and Ensemble Methods. FUW Trends in Science & Technology Journal. 3(2A); 518-522, Published by Federal University Wukari, Taraba State, Nigeria