Software defect prediction (SDP) is an essential task in software engineering for identifying defective modules early in the development process, thereby improving software quality and reducing maintenance costs. Existing SDP models often face a number of difficulties despite notable improvements, such as imbalanced datasets and the inability to capture the complex relationships within the code. In this manuscript, a software defect prediction using edge feature with self-attention-based cycle-consistent generative adversarial network optimized by the pelican optimization algorithm (SDP-ESA-CycleGAN-POA) is proposed to enhance the defect prediction by capturing semantic features within the software code. The proposed methodutilizesabstractsyntaxtree(AS-Tree)tokensanddensevectortransformationthrough word embedding to improve prediction accuracy. Then, e dge feature and self-attention-based cycle-consistent generative adversarial network (ESA-CycleGAN) predicts the data as buggy and clean. Finally, the pelican optimization algorithm (POA) is employed to enhance the weight parameters of the ESA-CycleGAN, leading to improved defect prediction accuracy. This research evaluates the ESA-CycleGAN model using the PROMISE dataset. The proposed SDP-ESA-CycleGAN-POA method achieves 21.57%, 23.41%, 16.10% and 18.73% higher accuracy compared with the existing models: a new method to SDP accuracy with machine learning (SDP-RF-LR), SDP using optimum trained convolutional neural network (SDP-OT-CNN), SDP utilizing intelligent ensemble-based method (SDP-RF-SVM-ANN) and SDP through neural network and feature selections (SDP-RBFNN-FS) respectively.
Download PDF: https://coloriage.eu.org/TBQ3jk
