Sofware Quality Prediction: An Investigation Based on Artificial Intelligence Techniques for Object-Oriented Applications

Abstract

AbstractExploring the relationship between Object Oriented software metrics and predicting the accuracy rate in defect prediction using the Chidamber & Kemerer (CK) metrics suite are the main purposes of this study. For these purposes, eleven machine learning (ML) techniques were analyzed to predict models and estimate reliability for eight open-source projects. Therefore, evaluating the relation between CK metrics and defect proneness was determined. Moreover, the performances of eleven machine learning techniques were compared to find the best technique for determining defect prone classes in Object Oriented software. The techniques were analyzed using Weka and RapidMiner tools and were validated using a tenfold cross-validation technique with different kernel types and iterations. Also, Bayesian belief’s network forms show which metrics are being the primary estimators for reliability. Due to the imbalanced nature of the dataset, receiver operating characteristic (ROC) analysis was used. The accuracy of the projects was evaluated in terms of precision, recall, accuracy, area under the curve (AUC) and mean absolute error (MAE). Our analyses have shown that Random Forest, Bagging and Nearest Neighbors techniques are good prediction models with AUC values. The least effective model is Support Vector Machines (SVM).KeywordsArtificial ıntelligence techniquesSoftware quality metricsSoftware reliabilityDefect predictionMachine learningObject Oriented metrics