Data Mining of Ambulatory Blood Pressure Monitoring

Abstract

The paper's aim is data mining of the recently published (2017) Ambulatory Blood Pressure Monitoring database, which covers 270 patients. Preprocessing, denoising, unsupervised Machine Learning (clustering), and feature selection were performed within Java-based software (WEKA 3-9-6). The Blood Pressure Load Level showed the best congruence with the obtained clusters. It also has the lowest noise among other levels of the database. The methods are designed to simplify database mining for clinicians largely unfamiliar with machine learning algorithms and benefit from simple algorithm approaches and visual aids output.