User-driven adaption in rule-based pattern recognition

Today, in software intensive projects a huge amount of the budget flows into the analysis of the already existing system. The reason for the high costs results mainly from the fact that analyses are often made manually or with automatic tool support, which is inappropriate for analyzing large systems. Semi-automatic analysis approaches usually use a notion of fuzziness to overcome this limitation, but inherit the problem of selecting appropriate initial values. In this paper we present an approach to adapt the initial values of our semi-automatic reverse engineering process. We provide the reverse engineer with accuracy information for results produced by a rule-based inference algorithm. Based on the changes of the results done by the reverse engineer we automatically adapt a credibility value of each rule, which previously has been used to compute the accuracy of the result. The adaption fits seamlessly into our overall analysis process. First tests show that it is suitable for the calibration of our fuzzyfied rule-based pattern recognition approach.