A Machine Learning Approach to Predicting High Blood Pressure Using Predictive Modeling on Local and Global Datasets to Enhance Patient Safety

Abstract

Hypertension remains a major global public health burden, contributing to cardiovascular disease and premature deaths.Despite advances in medical care, delayed diagnosis particularly in developing countries like Nigeria continues to undermine effective prevention and intervention strategies.Traditional approaches rely on periodic measurements, which often fail to capture early risk indicators/patient-specific factors.With the growing availability of large-scale clinical data, machine learning provides an opportunity to enhance predictive modeling for early detection.This study proposed a machine learning framework for predicting hypertension using both local (426 patient records from Federal Medical Centre, Yenagoa) and global datasets (174,982 instances from Kaggle).The dataset was preprocessed using python libraries.Four ML algorithms: Logistic Regression, Random Forest, K-Nearest Neighbor, and XGBoost were trained separately on different feature dimensions with evaluation metrics including accuracy, sensitivity, specificity, F1-score, and AUC-ROC.Results indicated that RF achieved ~99.95% accuracy on the global dataset, while XGB on local data attained ~98.84% with superior sensitivity in distinguishing high-risk categories.A prototype web app built from the best-performing model was successfully tested, showing strong clinical potential.The study highlights that using local and global datasets improved generalization, while ensemble models enhanced predictive reliability for early detection to improve patient safety.

Ähnliche Arbeiten