On a Neural-Fuzzy Technique with GA-Optimization for Software Quality Models

Managing software development and maintenance projects requires early knowledge about qual­ ity and effort needed for achieving this quality level. Effort estimations and staffing decisions are often based on productivity knowledge of the software development process. Quality-based pro­ ductivity management is introduced as one approach for achieving and using such process knowledge. Fuzzy rules are used as a basis for constructing quality models that can identify out­ Iying software components that might cause potential quality problems. This provides us with a fuzzy expert system tailored to the corresponding development environment. The value of any knowledge-based system is deterrnined by the accuracy and cost of such predictions which are used to improve life-cycle productivity. Using the self-Ieaming capability of neural networks, prestructured with genetic algorithms, the fuzzy rules can be automaticaJly generated from ex­ ample data to reduce the cost and improve the accuracy. The premises of the found rules consist of metrics drawing concJusions on the desired quality factors. The generated quality model - with respect to changes 2D provides both quality of fit (according to past data) and predictive accu­ racy (according to ongoing projects).