Generalized Canonical Correlation Based Bagging Ensembled Relevance Vector Machine Classifier for Software Quality Analysis

摘要

Early detection of defects helps to save cost and efforts. Few research works have been designed in existing works for analyzing the quality of software program using various machine learning techniques. However, the classification performance of existing work was lower which reduces the defect detection accuracy. In order to overcome the above existing issues, the Generalized Canonical Correlation Analysis based Bagging Ensembled Relevance Vector Machine Classification (GCCA-BERVMC) model is proposed. The GCCA-BERVMC model considers the number of source code lines from software program dataset as input. After that, the generalized Canonical Correlation Analysis (gCCA) algorithm designed in GCCA-BERVMC model selects the relevant features (i.e., code metrics) in order to improve the quality of software program. By applying Bagging Ensembled Relevance Vector Machine Classification (BERVMC) algorithm, the classification of program files is performed in GCCA-BERVMC model. The boosting algorithm creates 'n' number of weak learners to classify the input source code lines as normal or defected by analyzing the source codes and chosen metrics. After that, the weak learner's results are combined into strong classifier by using the majority votes. This helps for GCCA-BERVMC model to enhance the accuracy of defect prediction for software quality analysis with a lower amount of time. Experimental evaluation of GCCA-BERVMC model is conducted using metrics such as defect detection accuracy, false positive rate, and time complexity with respect to various software code sizes. The experimental result shows that the GCCA-BERVMC model is able to increase the defect detection accuracy and also minimizes the amount of time required for software quality analysis when compared to state-of-the-art works.