Software maintainability prediction of open source datasets using least squares support vector machines

Abstract

Software Maintainability (SM), being one of the priciest and tedious phases of any software development life cycle, has drawn the attention of various researchers over the years. SM measures the ease of carrying out maintenance activities such as repair and improvement of software code as per the changing needs of the customer and should be predicted well in advance. The current study implements the Least Squares Support Vector Machines (LS-SVM) algorithm for SM Prediction (SMP) on six open source datasets, namely Abdera, Ivy, jEdit, Log4j, Poi, and Rave. MAE, RMSE, and MMRE are considered as the prediction accuracy measures to evaluate the performance. Results indicate that LS-SVM is a potentially viable tool for predicting maintainability. Best results are obtained with jEdit dataset having minimum values for MAE, RMSE, and MMRE, i.e. 20.38, 46.97, and 1.02, respectively; jEdit having high component stability with lesser number of changes.

Subject Classification:

• Primary 68N30
• Secondary 68T01

Keywords:

• Machine Learning
• Software Maintainability Prediction
• LS-SVM
• Normalization