Software bug prediction using object-oriented metrics

Software quality is the fundamental requirement for a user, academia person, software developing organizations and researchers. In this paper a model for object-oriented Software Bug Prediction System (SBPS) has been developed. This model is capable of predicting the existence of bugs in a class if found, during software validation using metrics. The designed model forecasts the occurrences of bugs in a class when any new system is tested on it. For this experiment some open source similar types of defect datasets (projects) have been collected from Promise Software Engineering Repository. Some of these datasets have been selected for prediction of bugs, of which a few are not involved in model construction. First of all, we have formulated some hypotheses corresponding to each and every metric, and from metrics validation based on hypothesis basis finally 14 best suitable metrics have been selected for model creation. The Logistic Regression Classifier provides good accuracy among all classifiers. The proposed model is trained and tested on each of the validated dataset, including validated Combined Dataset separately too. The performance measure (accuracy) is computed in each case and finally it is found that the model provides overall averaged accuracy of 76.27%.