Construction of a hypertension risk prediction model based on deep learning

ZHANG Tianyu; XING Chunguo; MA Yuyang; SHENG Wenshun

doi:10.3969/j.issn.2096-501X.2026.02.017

Abstract：Objective To construct a hypertension risk prediction model and provide intelligent decision support for early screening,diagnosis,and treatment of hypertension in primary medical and health institutions.Methods Based on the Kaggle cardiovascular disease dataset,data standardization and visual analysis were performed to explore the distribution patterns of 11 features including age and blood pressure.A four-layer fully connected neural network was built using the Keras framework,with a Dropout layer embedded to prevent overfitting.Grid search was used to test parameter combinations of Adam/SGD optimizers,cross-entropy/mean squared error loss functions,learning rates ranging from 0.001 to 0.100,and batch sizes from 32 to 256.Five-fold cross-validation was employed to optimize the model structure.Results The constructed deep learning model showed stable performance on the test set.The binary cross-entropy loss function combined with the Adam optimizer achieved the optimal model,with a best training accuracy of 64% and a best validation accuracy of 70%.Parameter optimization experiments indicated that the combination of the Adam optimizer,0.010 learning rate,and 64 batch size achieved the optimal performance.This combination enabled the model to converge the fastest without obvious overfitting,significantly outperforming the SGD optimizer and other parameter combinations.Conclusions The neural network model established in this study can effectively identify high-risk populations for hypertension,providing an efficient auxiliary diagnostic tool for community hospitals and holding clinical application value for the secondary prevention of hypertension.