Software fault prediction is a process which predicts that the software modules are faulty or not by using the software metrics and some soft computing methods. Software metrics are divided into two main categories such as object-oriented and method-level metrics. While class relationships and dependencies are covered by object-oriented metrics, behaviors of the classes can be also measured by method-level metrics. Actually, the complementary relationship between these metric groups is focused in this study and different predictive models are built by using different parameter sets. Each parameter set includes some object-oriented and some method-level metrics. Furthermore, Mamdani style fuzzy inference system (FIS) is employed here to predict faultiness. In contrast to data-driven methods, FIS does not require historical or previous data for modeling. In fact, it is a rule-based approach and rules are extracted with the help of domain experts. In this study, the dataset which consists of the method-level and the class-level metrics' values that are collected from KC1 project of PROMISE repository is employed and most successful model whose performance is 0.8181 according to the evaluation criteria (the area under receiver operating characteristics (ROC) curve (AUC)) is built with the parameters of "coupling between object", "line of code" and, "cyclomatic complexity".

No abstract