6 Building a Prediction Model for Class Diagrams Maintainability

In the previous sections, we have found, by analysing the empirical data, that to some extent the metrics we presented for the structural complexity of UML class diagrams seem to be correlated with the class diagram maintainability sub-characteristics. This fact led us to think about building a prediction model for class diagram maintainability based on metric values. We have used all the metrics to build the prediction models because we considered it to be too premature to discard some metrics. Seeing the encouraging results obtained by the application of the Fuzzy Deformable Prototypes for building prediction models applied to different domains (Olivas et al., 2000; Olivas et al., 2002), we decided to use it for our purpose. For the sake of brevity we will not explain in depth all the steps of the prediction process. Further details can be found in Olivas (2000). The main goals of the prediction process, in our case, are:

1. The automatic search and extraction of fuzzy prototypes to characterise the class diagrams across its maintainability expressed as the time of maintenance. This step is called Fuzzy Prototypical Knowledge Discovery (FPKD) process.

2. The obtaining of a prediction model of the maintenance time of the class diagrams accomplishing the deformation of the previously discovered fuzzy prototypes.

To build the prediction model, we used the data obtained in the third experiment. Firstly, it was necessary to transform the data in order for it to be useful for the FPKD...