Two-Dimensional Multi-Deep CNNs for Accurate and Robust Software Defect Detection: A Performance-Driven Approach
Vinod Veeramachaneni, Pradeep Kumar Mallick, Subhashree Rout, Piyush Kumar Pareek · 2025
Software systems are growing in size and complexity. These properties make software defect prevention challenging. Thus, automatically predicting software module defects can help engineers conserve resources. Several ways have been presented to discover and fix these issues cheaply. However, these solutions may need a performance bump. This paper introduces the Multi-Deep Convolutional Neural Network to improve binary and multi-class software classification. Instead of one-dimensional CNNs, the proposed model captures n-gram characteristics inside and between datasets using a two-dimensional multi-scale convolutional technique. Advanced crow search optimisation algorithm (ACSOA) is used to optimally select model parameters. Besides Apache Active MQ, the study employed Eclipse data sets to evaluate the recommended deep learning method. F1-score, recall, accuracy, and precision were utilized to validate the model. The results show that the suggested strategy finds software vulnerabilities better and more reliably than current models.