<p>Software defect prediction aims to identify parts of a software project that might have issues early during development. This helps reduce the cost and effort required for testing. In this research, a new technique called Enhanced Moth-Flame Optimization (EMFO) is introduced to pick the most important features, along with an Entropy Centroid Support Vector Machine (EC-SVM) to make accurate predictions about defects. The process involves normalizing data with Z-score, selecting features using EMFO, and then transforming them into a hidden space where EC-SVM can classify the data. The method was tested on four NASA datasets—JM1, MC1, KC1, and PC2—and achieved a prediction accuracy of 96%, which is better than existing techniques. By focusing on the most important features, EMFO speeds up the training process and improves the separation between different classes. The results show that EMFO-EC-SVM is a powerful, adaptable, and efficient method. The study confirms that this approach works well and suggests it could be even more effective when used on larger or more unbalanced datasets in the future.</p>