Automated knowledge acquisition and application for software development projects

The application of empirical knowledge about the environment-dependent software development process is mostly based on heuristics. In this paper, we show how one can express these heuristics by using a tailored fuzzy expert system. Metrics are used as input, enabling a prediction for a related quality factor like correctness, defined as the inverse of criticality or error-proneness. By using genetic algorithms, we are able to extract the complete fuzzy expert system out of the available data of a finished project. We describe its application for the next project executed in the same development environment. As an example, we use complexity metrics which are used to predict the error-proneness of software modules. The feasibility and effectiveness of the approach is demonstrated with results from large switching system software projects. We present a summary of the lessons learned and give our ideas about further applications of the approach.