Children with congenital and acquired heart diseases are particularly at greater risk of suffering cardiac arrest. Despite significant progress in the development of resuscitation techniques and technologies, survival rates after cardiac arrest in this population remain stagnant. Predictive algorithms, possibly driven by machine learning, may have the ability to advance the incidence of cardiac arrest. The paper proposes a new hybrid deep learning model that combines convolutional neural networks (CNNs) and long short-term memory (LSTMs) to predict outcome results in pediatric heart disease. Because such a model can treat image and time-series data, including ECGs and echocardiograms, it can address some of the traditionally used method drawbacks for diagnosing diseases. Evaluated on diverse datasets, the hybrid CNN-LSTM model demonstrated superior performance in terms of accuracy, precision, and recall aspects performed better compared with the models using only CNN or LSTM. This study improved the detection of heart disease; it lowered false positives and negatives, thereby bringing down the high prevalence of misdiagnosis that happened in this clinical area. This brings out the prospect of combining machine learning models with clinical practices to identify CHD in children earlier. This model will be validated by future research, where it will be integrated into real-world health systems. Challenges that must be further understood to ensure robustness and scalability in a wide range of clinical settings are data imbalance, overfitting, and computational complexity.