Following the pandemic, the prevalence of heart diseases has been escalating at an alarming rate, making early detection crucial for effective management and health preservation. Deep learning-based models have emerged as powerful tools for early diagnosis, offering valuable insights into heart disease status. This study proposes a hierarchical learning approach for heart failure discernment, utilizing both patient text datasets and ECG images. Machine learning models are applied to text data, while a Convolutional Neural Network (CNN) model is employed for image data analysis. The comparison of prediction metrics demonstrates the efficacy of these models. Logistic Regression achieves an accuracy of 82.33%, Naïve Bayes 81.54%, SVM 80.33%, KNN 66.93%, Decision Tree 80.97%, Random Forest 88%, and XGBoost 85%. Notably, the CNN model outperforms all others, achieving an impressive accuracy of 97.7% on ECG images. These results underscore the significant potential of CNNs in advancing heart failure prognostication at its earliest stages, enabling timely intervention and promoting better health outcomes.