Effects of Mean Metric Value Over CK Metrics Distribution Towards Improved Software Fault Predictions

Abstract

Object Oriented software design metrics has already proven capability in assessing the overall quality of any object oriented software system. At the design level it is very much desirable to estimate software reliability, which is one of the major indicators of software quality. The reliability can also be predicted with help of identifying useful patterns and applying that knowledge in constructing the system in a more specified and reliable manner. Prediction of software fault at design level will also be helpful in reducing the overall development and maintenance cost. Authors have classified data on the basis of fault occurrence and identified some of the classification algorithm performance up to 97%. The classification is carried out using different classification techniques available in Waikato Environment for Knowledge Analysis (WEKA). Classifiers were applied over defect dataset collected from NASA promise repository for different versions of four systems namely jedit, tomact, xalan, and lucene. The defect data set consist of six metrics of CK metric suite as input set and fault as class variable. Outputs of different classifiers are discussed using measures produced by data mining tool WEKA. Authors found Naive Bayes classifier as one of the best classifiers in terms of classification accuracy. Results show that if overall distribution of CK metrics is as per proposed Mean Metric Value (MMV), the probability of overall fault occurrence can be predicted under consideration of lower standard deviation values with respect to given metric values.