Fuzzy inferencing to identify degree of interaction in the development of fault prediction models

The software fault prediction models, based on different modeling techniques have been extensively researched to improve software quality for the last three decades. Out of the analytical techniques used by the researchers, fuzzy modeling and its variants are bringing out a major share of the attention of research communities. In this work, we demonstrate the models developed through data driven fuzzy inference system. A comprehensive set of rules induced by such an inference system, followed by a simplification process provides deeper insight into the linguistically identified level of interaction. This work makes use of a publicly available data repository for four software modules, advocating the consideration of compound effects in the model development, especially in the area of software measurement.One related objective is the identification of influential metrics in the development of fault prediction models. A fuzzy rule intrinsically represents a form of interaction between fuzzified inputs. Analysis of these rules establishes that Low and NOT (High) level of inheritance based metrics significantly contributes to the F-measure estimate of the model. Further, the Lack of Cohesion of Methods (LCOM) metric was found insignificant in this empirical study.