An investigation of cross-project learning in online just-in-time software defect prediction

Defect Prediction Using Akaike and Bayesian Information Criterion

“…They found that supervised models are more promising than unsupervised methods. Tabassum et al [12] worked on finding that to what extent cross-project data is useful. Is it useful for just starting phase of the new project or it can also work when the project is old?…”

Defect Identification, Categorization, and Repair: Better Together

“…Besides, Wan et al [39] discussed the drawbacks of existing defect prediction tools and highlighted future research directions through literature review and a survey of practitioners. After that, Tabassum et al [36] conducted a study of JIT defect prediction in realistic online learning scenarios and concluded that the model trained with both within and cross-project data can outperform the model trained with within-project data only. Recently, Zeng et al [48] revisited the deep learning based JIT defect prediction models and found that deep learning based approaches may not work better than simplistic model LApredict they proposed.…”

A Novel Approach to Address External Validity Issues in Fault Prediction Using Bandit Algorithms

“…Online learning: Bandit algorithms can be considered as an online learning method [7]. Several online learning methods have been applied to fault prediction in the past (e.g., [17], [19]). However, the assumption in the past studies is that fault prediction models should be rebuilt continuously as prediction targets of software systems might change over time.…”

“…Tabassum et al [17] applied online learning to Just-In-Time software defect prediction models in order to compare the outcomes of three different proposed methods. Although the authors used majority voting (simply by counting the majority, and not the weighted majority [13]), their approach aimed at building multiple prediction models repeatedly, based on data points which are sequentially added to the models.…”

“…That is, they did not consider the accuracy of prediction models to optimize the prediction online. The study [17] assumes that the prediction models are not only used in the testing phase but also in the whole software development process. The study used datasets that were collected from a company from a period of 9-10 months, and other datasets that were collected from open source projects for a period of 6-14 years.…”