A practical method for the software fault-prediction
Zhan Li, Marek Reformat · 2007
In the paper, a novel machine learning method, SimBoost, is proposed to handle the software fault-prediction problem when highly skewed datasets are used. Although the method, proved by empirical results, can make the datasets much more balanced, the accuracy of the prediction is still not satisfactory. Therefore, a fuzzy-based representation of the software module fault state has been presented instead of the original faulty/non-faulty one. Several experiments were conducted using datasets from NASA Metrics Data Program. The discussion of the results of experiments is provided.