Software Defect Prediction with Design Metrics on NASA Dataset - Case Study Results

Within the scope of the case study, different ML-based models were constructed and applied on NASA dataset by using WEKA tool with the version of 3.8.5. The classifiers used for the experiments are summarized below: Artificial Neural Network (ANN): weka.classifiers.functions.MultilayerPerceptron Bayesian Belief Network (BBN): weka.classifiers.bayes.BayesNet Decision Tree (DT): weka.classifiers.trees.REPTree Fuzzy Rule Based (FRBC): weka.classifiers.rules.MultiObjectiveEvolutionaryFuzzyClassifier Logistic Regressin (LogR): weka.classifiers.functions.SimpleLogistic Naïve Bayes (NB): weka.classifiers.bayes.NaiveBayes Support Vector Machines (SVM): weka.classifiers. functions.LibLINEAR A 10-fold cross-validation approach was adopted in the training and testing stages of the classifiers. Cross-validation operations were run 10 times for different random segments, resulting in a total of 100 iterations. We reported both AUC and time-to-train values in order to benchmark the classifiers with regard to our research questions. The resulted values can be accessed through txt files