Software defect prediction plays a crucial role in ensuring the quality and reliability of software systems. In this paper, we propose a novel hybrid approach to software defect prediction that integrates a feature selection method with an ensemble machine learning model to enhance prediction accuracy. Our proposed method combines a two-stage feature selection process using filter and wrapper techniques, followed by the application of an ensemble learning algorithm that merges decision trees, support vector machines, and gradient boosting. This approach aims to address key challenges such as data imbalance and feature redundancy, which often affect the performance of defect prediction models. To evaluate the performance of the proposed method, experiments were conducted on publicly available software defect datasets, including the NASA MDP and PROMISE datasets. The results demonstrated that our hybrid model outperforms traditional machine learning algorithms, achieving a defect prediction accuracy of 94.3%, a precision of 92.1%, and an F1-score of 93.2%. Additionally, our method showed significant improvements in handling imbalanced data by reducing false-positive rates and improving recall. A comparative analysis with state-of-the-art approaches from the literature, including KAEA, metaheuristic optimization, and deep neural networks, revealed that the proposed hybrid model achieves superior performance in both cross-project and within-project defect prediction.