In modern software development, the early detection of defects is critical for ensuring the reliability and quality of software systems. However, software defect prediction (SDP) poses significant challenges, mainly when dealing with heterogeneous datasets comprising different feature types. Heterogeneity in the data can lead to class imbalance and hinder the performance of traditional SDP models. The proposed model integrates Generative Adversarial Networks (GAN) with Support Vector Machines (SVM) to predict software defects in heterogeneous datasets. The methodology involves several key steps: First, GANs are used to generate synthetic data samples to address class imbalance issues inherent in heterogeneous datasets. The synthetic, and original data are then utilized to train an SVM classifier, which is well-suited for handling complex data structures and providing robust predictions. The model’s performance is evaluated using metrics such as precision and recall demonstrating its effectiveness in software defect prediction. The experimental results indicate that the GAN-SVM approach effectively manages heterogeneous datasets, significantly improving the precision and recall rates compared to traditional SDP models. Specifically, the model achieved a precision of 98.00% for the CM1 dataset and 99.00% for the KC1 dataset, along with recall rates of 98.00% and 99.00% for CM1 and KC1, respectively. These findings underscore the GAN-SVM model’s robustness and superior performance in predicting software defects across diverse datasets. The proposed model offers significant improvements in precision and recall, providing valuable insights into the potential of combining these techniques for better defect prediction in software systems. Implementing GAN-SVM in software development pipelines could markedly improve defect prediction, aiding software engineers in maintaining high-quality systems.