Predicting the Onset of Hypertension Using Deep Learning Models in the Copperbelt Province of Zambia

Abstract - Hypertension continues to be a significant health challenge in Zambia, particularly in the Copperbelt Province, where many cases are diagnosed late and often advance to severe complications. This research aimed to design an artificial intelligence–driven system capable of predicting hypertension at earlier stages, thereby supporting preventive healthcare in resource-constrained settings. The study adopted a flexible, mixed-methods design that combined publicly available health datasets with professional insights from local medical practitioners. Predictive variables, including blood pressure readings, cholesterol levels, body mass index, and heart rate, were utilized to train various deep learning models. The approaches tested included Deep Neural Networks, Convolutional Neural Networks, and Recurrent Neural Networks enhanced with Long Short-Term Memory capabilities. Model performance was assessed using widely accepted evaluation measures, namely accuracy, recall, precision, the F1 measure, and the area under the curve. The findings indicated that the optimized Convolutional Neural Network achieved an accuracy level of slightly above 85 percent. In comparison, the Long Short-Term Memory model produced an accuracy of eighty-three percent, with a recall rate exceeding ninety percent in detecting hypertensive cases. To ensure the system was practical for end-users, it incorporated a user-friendly interface developed with Python Tkinter and Jupyter Notebook, enabling real-time prediction and reporting. Its modular server-client architecture enhanced both scalability and security, while model interpretability was supported through visualization techniques such as gradient-based mapping. The research also highlighted several challenges, including the shortage of structured local datasets, insufficient computing resources, and limited knowledge of artificial intelligence within the health sector. Despite these obstacles, the research demonstrated that tailored deep learning applications can strengthen public health decision-making in Zambia and provide a foundation for the development of future data-driven medical solutions, as exemplified by the prototype system developed.  Keywords - Deep Learning, Artificial Intelligence, Hypertension Prediction, Convolutional Neural Net, and Long Short-Term Memory.