Effort-Aware semi-Supervised just-in-Time defect prediction

•This is the first attempt to study the predictive effectiveness of sample-based semi-supervised models for effort-aware just-in-time defect prediction and an effort-aware tri-training method is proposed.•An in-depth evaluation on existing supervised and unsupervised models with time-wise cross-validation and cross-validation for various percentages of labeled data has been performed.•According to the results from a large-scale experimental comparison, the effectiveness of proposed model is verified even if the labeled data is insufficient.•The advantage of semi-supervised models is revealed as it can effectively combine the advantages of supervised and unsupervised models.