Evaluation of Training Methods for Conditioning of Fuzzy Based Maintainability Metric

The software maintainability can be ensured by carefully control of its software development process. An early measurement of maintainability starting from design phase is always desirable to produce maintainable software. Some of the researchers have tried to use soft computing techniques to measure maintainability. In spite of their reported validations, these models are not calibrated and no attention has been paid to evaluate and improve the stability of these methods. An attempt has been made in this paper to evaluate and compare several methodologies for improving the numerical stability of a fuzzy logic based maintainability metrics system. Tuning of fuzzy system parameters is carried out using genetic algorithm with system condition number as objective function for optimization. A number of alternates are considered, in which training data sets are generated using different methods and these sets are used to evaluate objective functions in GA and accordingly fuzzy parameters are tuned. In order to show the advantage of such stability improvement, real projects’ maintainability data is used and our study indicates that fuzzy model performance gets increased after conditioning. Key-Words: Leave software maintainability, soft computing, software metrics.