A cross-item defect prediction method using adversarial learning

• Published in: Journal of Autonomous Intelligence
• Main Author: Ma J.; Zain J.M.; Wang D.; Shi J.
• Format: Article
• Language: English
• Published: Frontier Scientific Publishing 2024
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85185701078&doi=10.32629%2fjai.v7i4.641&partnerID=40&md5=fec2051e314ae964c3211033579fd9d5

In today’s information society, the rapid growth of information technology has resulted in software products being integrated into every aspect of people’s lives. Consequently, the ability to accurately identify software modules that may cause problems within a specified time frame has become crucial for determining software development progress. This is because ensuring software dependability is a critical component of software development. In this paper, the enhanced abstract continuous generative adversarial network (AC-GAN) technique covers data processing and model construction. Three levels: (1) convert the code of the source project and the target project into the form of an abstract syntax tree (Unified Abstract Syntax Tree, UAST), then traverse the abstract syntax tree in a depth-first manner to obtain a node sequence, and then use continuous recursion to replace the nodes in the node sequence; (2) the processed numerical vectors are sent to the GAN-based model; (3) the GAN-based model generates the final word vectors. The network structure model is utilized for feature extraction and data transfer, and a binary classifier is then employed to determine if the target item code file is flawed. 15 sets of source-target item pairings are used to evaluate the AC-GAN approach. The experimental findings demonstrate the usefulness of the technique. © 2024 by author(s).