Convolution neural network-AlexNet with gazelle optimization algorithm-based software defect prediction

Convolution neural network-AlexNet with gazelle optimization algorithm-based software defect prediction

연구 분야: Verification

학회: Knowledge and Information Systems

Nowadays, the size and complexity of software systems have increased dramatically. Software defects are very challenging to prevent because of these characteristics. Therefore, developers may be able to better allocate their limited resources by predicting the number of defects in software modules automatically. There are various approaches presented for identifying and fixing such problems, but none of these give sufficient results. To address these, this paper proposes convolution neural network-AlexNet with gazelle optimization algorithm-based software defect prediction (SWDP-CNN-AlexNet-GAOA). Here, NASA software defect prediction dataset is used. The feature normalization ensures that all features contribute equally to the model. Without normalization, features with larger numerical values would dominate the learning process. Then, software defects are predicted using convolution neural network (CNN)-AlexNet. Finally, gazelle optimization algorithm (GAOA) is proposed to optimize the parameters of CNN-AlexNet. Simulation proves that the SWDP-CNN-AlexNet-GAOA method outperforms existing models. The proposed SWDP-CNN-AlexNet-GAOA approach attains 3.88%, 5.75%, and 4.94% better accuracy and 6.25%, 5.91%, and 11.28% better F-measure compared with the existing methods, like software defect prediction using enhanced CNN (SWDP-EN-CNN), software defect prediction using hybrid swarm intelligence and deep learning (SWDP-HS-DL), and software defect prediction under ant colony optimization (SWDP-ACO), respectively.