A Framework for Analyzing Software Quality using Hierarchical Clustering

Fault proneness data available in the early software life cycle from previous releases or similar kind of projects will aid in improving software quality estimations. Various techniques have been proposed in the literature which includes statistical method, machine learning methods, neural network techniques and clustering techniques for the prediction of faulty and non faulty modules in the project. In this study, Hierarchical clustering algorithm is being trained and tested with lifecycle data collected from NASA projects namely, CM1, PC1 and JM1 as predictive models. These predictive models contain requirement metrics and static code metrics. We have combined requirement metric model with static code metric model to get fusion metric model. Further we have investigated that which of the three prediction models is found to be the best prediction model on the basis of fault detection. The basic hypothesis of software quality estimation is that automatic quality prediction models enable verification experts to concentrate their attention and resources at problem areas of the system under development. The proposed approach has been implemented in MATLAB 7.4. The results show that when all the prediction techniques are evaluated, the best prediction model is found to be the fusion metric model. This proposed model is also compared with other quality models available in the literature and is found to be efficient for predicting faulty modules. Keywordsclustering; fault prediction; defect data; software quality; metrics.