A Hybrid Approach Entropy-TOPSIS for the Selection of Machine Learning Classifiers for Software Defect Prediction

The software is developed with much more complex functionalities to meet user requirements. Therefore, it is more vulnerable and subject to defects during the software development life cycle (SDLC). There are many efforts to avoid and re-duce the number of defects, ensure good performance, and achieve defect-free software before releasing it to the market. To minimize the effort and optimize testing using Machine Learning, classifier models can be created to classify and predict which modules of the developed software may or may not be more prone to defects. There is a wide variety of classifier models; however, there is no classifier model that performs better than another in general terms. Various performance metrics can be used using multiple historical data sets to compare and evaluate classifier models. To select the best classifier model, a hybrid approach combining two multicriteria decision making (MCDM) methods, Entropy and TOPSIS, is proposed. Entropy is used to calculate the criteria weights, and TOPSIS compares and ranks the alternatives. The results show that the proposed hybrid method can make the distribution of weights more reasonable and the selection of other options more efficient.