Deep Learning-Based Cross-Project Defect Prediction: A Comprehensive Survey

Software defect prediction has become an indispensable strategy for preserving system quality and optimizing maintenance efficiency in modern software engineering. However, data scarcity in new projects often hampers the effectiveness of defect prediction models. cross-project defect prediction (CPDP) addresses this challenge by transferring knowledge from source projects to target projects. Recent advances in machine learning, particularly deep learning, have shown significant potential in improving the accuracy and effectiveness of CPDP models. This paper presents the first comprehensive survey of deep learning-based CPDP (DL-CPDP) research, covering both homogeneous CPDP and heterogeneous CPDP. Specifically, we explore the application of deep learning architectures, machine learning paradigms, dataset utilization, evaluation metrics, and reproducibility practices. Our findings reveal key trends, such as the diversification of feature representations, the growing complexity of hybrid models, and the increasing adoption of semi-supervised learning techniques. Furthermore, the survey underscores the rising emphasis on reproducibility and open science, highlighting their role in advancing the field. This paper paves the way for more effective and innovative applications of deep learning in CPDP.