A Step Towards Software Reliability Prediction Using Support Vector Machine (SVM)

摘要

The increased demand of software in real life and safety critical systems has intensified the pressure on project managers to deliver more reliable software. In this article, the authors designed a support vector machine (SVM) model for predicting software reliability. The proposed model has been validated using various parameters applied on 16 software life-cycle empirical datasets extracted from the Data and Analysis Center for Software (DACS). The performance of the SVM model is measured and compared with other commonly used techniques of ANNs (radial basis function network, multilayer perceptron) and decision trees (REPTree and decision stump). The goal of their study is to facilitate software managers to determine the software release instance within time and budget constraint. Thus, software developers can predict the remaining number of faults in the software during the testing phase in order to improve the quality of software system. From rigorous experiments conducted, the authors observed that performance of the SVM model is better and it outperforms the models predicted using ANNs-based and DTs-based prediction models for software reliability in terms of MAE, RMSE, and RAE. Finally, they conclude that the SVM model is more reliable, more accurate with better capability of generalization, and less dependent on the sample data size.