Software defect prediction is important for improving the quality, reliability, and efficiency of software development processes. In this research, a defect prediction framework based on Convolutional Neural Networks (CNNs) with a data balancing method through the Synthetic Minority Over-sampling Technique (SMOTE) is proposed. The approach utilizes static code metrics from NASA’s MDP datasets and consists of some important steps such as data preprocessing, normalization, and reshaping to improve model performance. This method efficiently handles issues related to unbalanced datasets, thus improving prediction accuracy. It is especially useful for large software projects by facilitating early defect detection, which results in lower development costs, fewer post-release problems, and more reliable software. Through timely detection of defects, this framework helps in effective resource utilization and facilitates the provision of high-quality, robust, and efficient software systems. Overall, the combination of CNNs and SMOTE is a breakthrough in software defect prediction approaches.