An Empirical Validation of Object-Oriented Design Metrics for Fault Prediction

Abstract

Problem Statement: Object-oriented design has become a dominant method in software industry and many design metrics of object-oriented programs have been proposed for quality prediction, but there is no well-accepted statement on how significant those metrics are. In this study, empirical analysis is carried out to validate object-oriented design metrics for defects estimation. Approach: The Chidamber and Kemerer metrics suite is adopted to estimate the number of defects in the programs, which are extracted from a public NASA data set. The techniques involved are statistical analysis and neuro-fuzzy approach. Results: The results indicate that SLOC, WMC, CBO and RFC are reliable metrics for defect estimation. Overall, SLOC imposes most significant impact on the number of defects. Conclusions/Recommendations: The design metrics are closely related to the number of defects in OO classes, but we can not jump to a conclusion by using one analysis technique. We recommend using neuro-fuzzy approach together with statistical techniques to reveal the relationship between metrics and dependent variables, and the correlations among those metrics also have to be considered.

Faster issue resolution with higher technical quality of software

“…Interestingly, the often-used McCabe complexity (1976) is not included in this analysis. A number of later works have replicated the verification of the Chidamber and Kemerer metrics, using various techniques such as threshold models (Benlarbi et al 2000) and neural networks (Xu et al 2008). Zimmermann et al (2009) examine the transferability of bug prediction models between projects.…”

Study of Object Oriented Analysis and Design Approach

“…A similar study by (Xu et al, 2008) focused only on those metrics that are available at the design stage. The measures involved two characteristics of OO design classes, coupling and inheritance (briefly explained above).…”

“…Data were collected on faults reported by users of both versions so that classes could be identified as either faulty or not. Design metrics were applied to all classes in both versions to find the relationships between measures of coupling and inheritance and fault-proneness of the classes (Xu et al, 2008).…”

Adaptive Software Fault Prediction Approach Using Object-Oriented Metrics

“…We were quite surprised, that otherwise metric identified as good predictor in previous studies [6,22,62,46], RFC has not exhibited even a single good fit of data value produced by the HL test as an inferential goodness-of-fit test. In the wake of our findings, we select three metrics as best candidates for our logistic regression and ultimately the development of our fault prediction models: CBO, WMC, and LCOM.…”

“…Xu et al [62] utilize linear regression and a neuro-fuzzy approach to validate relationships between CK metrics suite and number of faults in OO classes. Investigated applications belong to the public NASA data set repository and are implemented in C++.…”