A Predictive Modeling approach for Early Detection of Hypertension

Abstract

Despite being one of the major modifiable risk factors for cardiovascular disease, stroke and premature death globally, a significant proportion of those affected are not detected or managed well. This paper presents the design and development of a real-time predictive system for early detection of hypertension risk by continuously streamed physiological parameters such as BP, HR, Peripheral oxygen saturation (SpO2), and body temperature. The engineered features such as pulse pressure, mean arterial pressure (MAP), and rate-pressure product (RPP) are calculated from raw sensor readings and then used to train supervised machine learning classifiers namely logistic regression, random forest, support vector machine (SVM), and gradient boosting. A labelled set of 12,000 physiological signals were used to train and validate models through stratified 5-fold cross-validation and hyper-parameter optimization via grid-search. The models (LDA, SVM, NB, RF) performed well, with the RF model chosen as the deployed model, yielding the highest test accuracy of 99.8%, precision of 100%, recall of 99.2%, F1-score of 0.996 and ROC-AUC of 1.000. The trained model was embedded in a real-time dashboard that feeds streaming sensor data into the model for real-time risk calculation and alerts the clinician when the risk for hypertension is greater than a preset limit, with a retraining loop using feedback to improve the model. The results show that the use of cheap IoT sensing in combination with light weight machine learning for continuous, real-time hypertension screening and early-warning support is feasible, both in clinical settings and in home monitoring.