An Evaluation and Selection of Machine Learning Models for Blood Pressure Prediction

Abstract

Blood pressure (BP) prediction using machine learning (ML) algorithms has emerged as a critical area of research in the field of healthcare.  ML methods provide particular advantages by using the power of large datasets.  Firstly, they enable the creation of tailored predictive models that can take into account a wide range of individual characteristics such as medical history and lifestyle.  When carefully trained and validated, these models can detect high blood (BP) pressure at an early stage, allowing for preventive interventions and customized healthcare regimens.  Furthermore, the use of ML in BP prediction can help to reduce healthcare costs by optimizing resource allocation and also offers a viable route for improving the precision of healthcare interventions.  This study used an ML model to predict future fluctuations of an individual’s BP using their future calendar events.  The study was done in Uasin-Gishu County, Kenya.  The design science method was employed for the study.  The data was collected using a smartwatch, which collected the BP and heart rate, and a smartphone application which collected the individuals' moods, activities, and calendar events.  The algorithms that were selected and evaluated for the predictive ML models are; Lasso Regression, Linear Regression, ElasticNet, K-Nearest Neighbors (KNN), Decision Tree Regressor, and Gradient Boosting Regressor (GBR).  The Holdout method’s test data set, R-squared (R2), and Mean Squared Error (MSE) were used to evaluate the models.  The GBR predictive model was the best performing out of the selected models and was implemented using the Iterative and Incremental Development Model.  The GBR model gave an MSE score of 0.182 and R² score of 0. 992. A two-sample T-test was also conducted giving a T-statistic of 0.047 and a P-value of 0.963. These scores depict the GBRs’ good performance in predicting an individual’s BP using future planned activities.