Exploring HDBSCAN for Effort-Aware Defect Prediction: A Density-Based Clustering Method

This study explores the application of HDBSCAN, a density-based unsupervised clustering method, in effort-aware defect prediction (EADP). Unlike traditional methods, HDBSCAN dynamically identifies defect clusters and removes noise points without requiring predefined parameters. Experiments on the PROMISE dataset demonstrate that HDBSCAN outperforms several supervised and unsupervised methods in metrics such as IFA and PMI@20%, reducing testing resource waste and improving early defect detection. However, its performance on pof B@20% is slightly inferior to certain state-of-the-art supervised methods. These results indicate that HDBSCAN is an efficient alternative for EADP and suitable for defect prediction tasks.