Related Experiment Video
Updated: May 10, 2025
08:12
Calculating Heart Rate Variability from ECG Data from Youth with Cerebral Palsy During Active Video Game Sessions
Published on: June 5, 2019
19.7K
Development of Enhanced Machine Learning Models for Predicting Type 2 Diabetes Mellitus Using Heart Rate Variability:
Vinni S Fengade1, Hira Swati1, Manoj Chandak1
1Computer Science and Engineering, Shri Ramdeobaba College of Engineering and Management, Nagpur, IND.
Cureus
|April 21, 2025
Summary
Machine learning models analyzing heart rate variability (HRV) show promise for non-invasive type 2 diabetes mellitus (T2DM) screening. Optimized models achieved high accuracy, offering a scalable alternative to traditional methods.
Area of Science:
- Cardiology
- Biomedical Engineering
- Data Science
Background:
- Type 2 Diabetes Mellitus (T2DM) is a growing global health concern requiring early, non-invasive detection methods.
- Current screening methods (OGTT, HbA1c, fasting plasma glucose) have limitations including invasiveness and cost.
- Heart Rate Variability (HRV) analysis combined with Machine Learning (ML) presents a potential non-invasive screening approach, addressing generalizability issues of prior HRV-based ML models.
Purpose of the Study:
- To develop and validate ML models using diverse HRV features (time-domain, frequency-domain, nonlinear) for enhanced T2DM prediction.
- To evaluate the performance of these developed ML models against existing ML models for T2DM classification.
- To identify key HRV parameters crucial for accurate diabetes prediction.
Main Methods:
- Retrospective analysis of ECG recordings from 519 individuals (261 T2DM patients, 258 controls).
- Extraction of time-domain, frequency-domain, and nonlinear HRV features following established guidelines.
- Training and assessment of various ML models (Logistic Regression, KNN, Random Forest, Gradient Boosting, XGBoost, LightGBM, CatBoost, AdaBoost) using an 80:20 train-test split, with hyperparameter optimization via GridSearchCV.
- Performance evaluation using accuracy, AUC, sensitivity, and specificity.
Main Results:
- Significant differences in time-domain, frequency-domain, and nonlinear HRV parameters were observed between T2DM patients and controls (p<0.001).
- Ensemble models, particularly CatBoost (91.2% accuracy, 0.91 AUC) and LightGBM, showed superior predictive performance.
- K-Nearest Neighbors (KNN) achieved the highest accuracy (98.2%) and AUC (0.99) after hyperparameter tuning, followed closely by Random Forest.
- SD2, SDRR, and CVRR were identified as the most significant HRV features for T2DM prediction.
Conclusions:
- HRV-based ML models are validated as effective tools for non-invasive T2DM prediction.
- Ensemble models like CatBoost and LightGBM demonstrate superior performance compared to previous ML models.
- Optimized ML models integrated with wearable technology offer a scalable, affordable, and non-invasive screening solution for diabetes, warranting further investigation in wearable-based monitoring and multimodal AI.
