Photoplethysmography-based non-invasive blood pressure monitoring via ensemble model and imbalanced dataset processing

Qianyu Liu, Chaojie Yang, Sen Yang, Chiew Foong Kwong, Jing Wang, Ning Zhou

Research output: Journal PublicationArticlepeer-review

Abstract

Photoplethysmography, a widely embraced tool for non-invasive blood pressure (BP) monitoring, has demonstrated potential in BP prediction, especially when machine learning techniques are involved. However, predictions with a singular model often fall short in terms of accuracy. In order to counter this issue, we propose an innovative ensemble model that utilizes Light Gradient Boosting Machine (LightGBM) as the base estimator for predicting systolic and diastolic BP. This study included 115 women and 104 men, with experimental results indicating mean absolute errors of 5.63 mmHg and 9.36 mmHg for diastolic and systolic BP, in line with level B and C standards set by the British Hypertension Society. Additionally, our research confronts data imbalance in medical research which can detrimentally affect classification. Here we demonstrate an effective use for the Synthetic Minority Over-sampling Technique (SMOTE) with three nearest neighbors for handling moderate imbalanced datasets. The application of this method outperformed other methods in the field, achieving an F1 score of 81.6% and an AUC value of 0.895, emphasizing the potential value of SMOTE for addressing imbalanced datasets in medical research.

Original languageEnglish
JournalPhysical and Engineering Sciences in Medicine
DOIs
Publication statusAccepted/In press - 2024

Keywords

  • Data imbalance
  • Hypertension
  • Machine learning
  • PPG

ASJC Scopus subject areas

  • Biotechnology
  • Radiological and Ultrasound Technology
  • Biophysics
  • Biomedical Engineering
  • Instrumentation
  • Radiology Nuclear Medicine and imaging

Fingerprint

Dive into the research topics of 'Photoplethysmography-based non-invasive blood pressure monitoring via ensemble model and imbalanced dataset processing'. Together they form a unique fingerprint.

Cite this