Predicting the Risk of Hypertension Based on Several Easy-to-Collect Risk Factors: A Machine Learning Method

• DOI: 10.3389/fpubh.2021.619429
• Publication Date: 2021-09-24T16:28:06Z
• Authors: Huanhuan Zhao; Xiaoyu Zhang; Yang Xu; Lisheng Gao; Zuchang Ma; Yining Sun; Weimin Wang
• Keywords: Univariate; Predictive modelling

Hypertension is a widespread chronic disease. Risk prediction of hypertension an intervention that contributes to the early prevention and management hypertension. The implementation such requires effective easy-to-implement risk model. This study evaluated compared performance four machine learning algorithms on predicting based easy-to-collect factors. A dataset 29,700 samples collected through physical examination was used for model training testing. Firstly, we identified factors hypertension, univariate logistic regression analysis. Then, selected features, 10-fold cross-validation utilized optimize models, random forest (RF), CatBoost, MLP neural network (LR), find best hyper-parameters set. Finally, models by AUC, accuracy, sensitivity specificity test experimental results showed RF outperformed other three achieved AUC 0.92, accuracy 0.82, 0.83 0.81. In addition, Body Mass Index (BMI), age, family history waist circumference (WC) are primary These findings reveal it feasible use algorithms, especially RF, predict without clinical or genetic data. technique can provide non-invasive economical way in large population.