Related Experiment Video
Updated: May 29, 2025
08:23
Characterization of Sickling During Controlled Automated Deoxygenation with Oxygen Gradient Ektacytometry
Published on: November 5, 2019
9.7K
Estimation of Hematocrit Volume Using Blood Glucose Concentration through Extreme Gradient Boosting Regressor Machine
Kirti Sharma1, Pawan K Tiwari1, S K Sinha1
1Department of Physics, Birla Institute of Technology, Mesra, Ranchi 835215, India.
Journal of Chemical Information and Modeling
|February 5, 2025
Summary
This study predicts hematocrit volume using machine learning models based on glucose concentration. The XGBoost model showed promising accuracy for biomedical signal processing applications.
Area of Science:
- Biomedical Signal Processing
- Machine Learning in Healthcare
- Diabetes Management
Background:
- Lifestyle diseases like diabetes significantly impact physiological metabolism and are linked to numerous other health conditions.
- Effective health management for diabetes involves diet, exercise, and regular monitoring of blood glucose and hematocrit levels.
- Current methods for monitoring hematocrit volume can be invasive or require specialized equipment.
Purpose of the Study:
- To develop and evaluate machine learning models for estimating hematocrit volume from glucose concentration data.
- To explore the efficacy of various regression models, including Linear Regression, Support Vector Regressor, Decision Tree, Random Forest Regressor, Artificial Neural Network, and Extreme Gradient Boosting Regressor.
- To enhance diagnostic capabilities in biomedical signal processing by providing a non-invasive method for hematocrit estimation.
Main Methods:
- Utilized amperometric signals from electrochemical glucose sensors to correlate glucose concentration with hematocrit volume.
- Trained and tested multiple machine learning models (LR, SVR, DT, RFR, ANN, XGBoost) using an 80% training and 20% testing dataset in Python.
- Evaluated model performance using R-squared, Mean Squared Error, and Root Mean Squared Error, with reliability assessed via relative error, K-fold cross-validation, and confidence interval analysis.
Main Results:
- The Extreme Gradient Boosting (XGBoost) regression model demonstrated superior performance compared to Linear Regression and Artificial Neural Network models.
- XGBoost achieved a 15% relative error between actual and predicted hematocrit values.
- The XGBoost model exhibited 68% accuracy with a 6% standard deviation, validated through 5-fold cross-validation.
Conclusions:
- Machine learning, particularly the XGBoost model, offers a viable approach for estimating hematocrit volume from glucose concentration data.
- The XGBoost model's performance, flexibility, and interpretability make it suitable for predictive biomedical analytics.
- This non-invasive method has the potential to improve diagnostic capabilities in managing conditions linked to diabetes and hematocrit levels.
