Answers are grounded in this paper and cite their sources.
Cross-Project Defect Prediction (CPDP) identifies defects in an unlabeled project (target) by using the historical data of other source projects. CPDP is particularly important, as manually identifying defects in newly developed software is both time-consuming and resource-intensive. There are various approaches for CPDP, with most focusing primarily on algorithmic development. However, traditional CPDP datasets suffer from two major challenges: the number of defective instances is often very limited, and the data distribution is highly imbalanced. As a result, the majority class tends to dominate the learning process. Synthetic Minority Oversampling Technique (SMOTE) and Adaptive Synthetic Sampling (ADASYN) are widely used approaches for solving class imbalance problem. However, these methods do not consider information from the target data during oversampling, and they are often susceptible to the influence of outliers, which may misguide synthetic sample generation. To address these challenges, we propose Outlier-Conscious Transfer-Oriented Minority Oversampling (OTOMO), a domain adaptation-based oversampling approach for CPDP. OTOMO incorporates target data distribution while reducing the impact of outliers, thereby generating more representative synthetic samples. Experimental results on four widely used benchmark CPDP datasets, namely NASA, RELINK, SOFTLAB and AEEEM, demonstrate that OTOMO achieves competitive or superior performance compared with existing oversampling techniques across multiple evaluation metrics such as ROC-AUC, G-mean, F1-score and MCC in two different classifiers Random Forest (RF) and Logistic Regression (LR).
No takes yet. Share an insight, caveat, or question.
Akhter et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: