Predicting defect modules in advance, software fault prediction plays an important role in improving software quality and reducing the cost of maintaining software. In this study we propose a hybrid deep learning model by combining CNN with LSTM networks and Dense layers to enhance predictive performance. This hybrid of CNNs, LSTMs and fully linked layers harnesses the spatial feature extraction capabilities of CNNs, the temporal dependence modeling of LSTMs and the abstraction ability of fully connected layers. Using a real-world software engineering dataset for model training and measuring, results are compared with independent CNN, RNN and Cascade models. The Performance is measured using evaluation metrics such as Accuracy, F1 Score, Precision, Recall, ROC AUC. Among these models, the proposed hybrid model achieves the highest accuracy of 91.4%, with an F1 score of 0.894, outperforming all baseline models. The findings demonstrate that the hybrid technique outperformed single techniques in terms of recognizing diverse trends of software metrics and boosting the reliability of fault prediction systems. These findings support the use of multi-branch deep learning architectures in complex software engineering tasks and pave the way for future refinement of more efficient defect detection approaches.