کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Non-Invasive Blood Glucose Estimation Using Data Trimming with PCA-Based Multi-Cycle Selection

نویسندگان :

Yeganeh Binafar¹ Arian Mesforoosh-M² Mohammad-R Akbarzadeh-T³ Farveh Daneshvarfard⁴

1- دانشگاه فردوسی مشهد 2- دانشگاه فردوسی مشهد 3- دانشگاه فردوسی مشهد 4- دانشگاه فردوسی مشهد

کلمات کلیدی :

Non-invasive glucose estimation،Photoplethysmography (PPG)،Biomedical signal processing،Data trimming،Principal component analysis (PCA)،Machine learning

چکیده :

Non-invasive blood glucose estimation using photoplethysmography (PPG) signals has gained significant attention due to its potential for continuous and cost-effective monitoring. However, cycle-to-cycle variability and morphological distortions in raw PPG signals can significantly degrade prediction accuracy. In this study, we propose a multi-cycle selection and dataset trimming framework that combines principal component analysis, K-means clustering, and reconstruction error to identify representative and exceptional cycles, reducing data redundancy while preserving critical morphological information. The trimmed dataset was used to train four regression models, namely Gradient Boosting, Extreme Gradient Boosting, Multi-Layer Perceptron, and K-Nearest Neighbors (KNN) for estimating reference glucose levels measured by a clinical glucometer. Experimental results show that the trimmed KNN model achieved the best performance with a Mean Absolute Error (MAE) of 14.15 mg/dL, Root Mean Square Error (RMSE) of 19.81 mg/dL, and Mean Absolute Relative Difference (MARD) of 11.89%. Clarke Error Grid analysis confirmed that 100% of predictions fell within clinically safe zones A and B. These results demonstrate that the proposed multi-cycle selection method effectively improves morphological consistency and non-invasive glucose estimation, providing an efficient alternative to complex deep learning models.