Abstract

The escalating prevalence of hypertension necessitates the development of sophisticated predictive mechanisms in healthcare. This study introduces a hybrid methodology synergizing machine learning and deep learning for feature augmentation and hypertension prognosis. We investigated five machine learning algorithms Random Forest (RF), Logistic Regression (LR), Decision Tree (DT), XGBoost (XGB), and Gradient Boosting (GB) to generate predictive features, which were subsequently integrated with the original dataset to train a Long Short-Term Memory (LSTM) network. Comparative analysis reveals that the Gradient Boosting-based Feature Prediction combined with LSTM (GB-based FP + LSTM) yields superior performance, achieving a peak accuracy of 98.48% and an equivalent F1-score. Conversely, the LR-based integration demonstrated the lowest accuracy at 89.39%. These results suggest that the proposed GB-based FP + LSTM framework offers a highly reliable diagnostic instrument, providing clinicians and policymakers with an effective pathway for early and accurate hypertension detection.