Value and Applicability of Academic Projects Defect Datasets in Cross-Project Software Defect Prediction

This paper presents a novel concept in cross-project software defect prediction. Defects in real world software systems are predicted, using defects data of academic projects as training data. Relevant training data are filtered using patterns by ordered projections algorithm by [14]. A satisfactory Area under curve(AUC) value of at least 0.7 and above is obtained for 12 out the 16 investigated real world software systems with Bayes Network classifier. This result is quite significant in the current scenario where many software startups are being launched by fresh graduates from Universities, particularly in India. At the beginning such startup companies do not have any defects data of real projects and hence defect datasets of academic projects collected at Universities can be really useful.