Machine Learning Approach to Predict Software Evolvability using Fuzzy Binary Trees

摘要

We present a technique to circumvent a major problem while implementing and applying techniques to build software quality estimation models, namely the use of rules with precise threshold values for metrics. We propose a machine learning approach that uses a fuzzy binary decision tree instead of the more classical binary decision tree. After presenting this novel approach, we describe an example application that investigates the stability of a class library interface, as we move form version to version of the library, using structural metrics as stability indicators. We conducted a study on different versions of a commercial C++ class library using both a standard binary tree approach and our fuzzy logic approach; the obtained results are very promising when comparing the two methods.