Deep Learning Algorithm for Hypertension Prediction using Hyperparameter Optimization via Latin Hypercube Sampling in High-Performance Computing Environments

Resumen

Hypertension is a chronic condition characterized by persistently elevated blood pressure. Due to its asymptomatic
nature in the early stages, it is frequently underdiagnosed, leading to diagnosis only after the occurrence of cardiovascular complications. This silent progression highlights the need for early diagnostic tools capable of identifying at-risk individuals
before irreversible damage occurs to target organs. To fill this clinical gap, this study proposes a Deep Learning Algorithm for
Hypertension Prediction based on clinical variables. A Hyperparameter Optimization process was conducted to define the architecture and adjust the predictive capacity. The search space was systematically explored using Latin Hypercube Sampling (LHS) to define the neuron configuration, learning rate, dropout, and L2 regularization. Due to the computational complexity of evaluating multiple neural network architectures through 10-fold Stratified Cross-Validation, the optimization pipeline was implemented in High-Performance Computing (HPC) environments. The model achieved an accuracy of 90.21% and an F1-Score of 90.19%, demonstrating the balance achieved between precision and recall.