NEURO FUZZY MODEL FOR SOFTWARE MAINTAINABILITY PREDICTION

[摘要] Component-Based Software Engineering (CBSE) is kind of approach to develop software systems by using existing components with the objective to improve software quality and productivity. However, after deployment on customer site, maintenance of Component Based Software (CBS) becomes an important issue and is difficult to perform and manage effectively. It faces challenges of selecting combinations of components, effective techniques of component storage in repository, creating components and integrating the selected components. Performing a good component selection plays a critical role in the success of the final system. The objective of our work is to determine the quality factors that influencing CBS maintenance costs  and to provide benchmark indications for designing efficient maintenance oriented software that contribute to the enhancement of business performance. In order to achieve this we have proposed a neuro-fuzzy model for maintainability prediction for component based system. Proposed model is trained with five inputs namely documentation quality, modifiability, integrability, testability and coupling between components to automatically predict software maintainability. Firstly, Fuzzy Logic is used here to construct set of rules (3 5i.e. 243 rules) and generate training and testing data sets. In our proposed work, we have developed neuro-fuzzy model with the data sets generated by the fuzzy logic to take the advantage of some of the desirable features of a neuro-fuzzy approach, such as learning ability and good interpretability. The model was trained well and predicts satisfactory results with MARE 22% and MRE 0.007%.

[学科分类] 建筑学

[关键词] Rule mining;non-random initialization;entropy based;classification rule