Innovation in Blood Sugar and Cholesterol Measurement Tools Based on Integrated Sensors for Non-Invasive Health Monitoring

Purpose: Non-invasive monitoring of blood glucose and cholesterol has become increasingly important due to the growing prevalence of diabetes and cardiovascular diseases and the limitations of conventional invasive measurement methods, which require blood sampling, cause discomfort, and increase the risk of infection. Although various non-invasive techniques have been proposed, most existing studies focus on a single biomarker or a single sensing modality, limiting their robustness and practical implementation for real-time health monitoring. This study aims to develop an Internet of Things (IoT)-based non-invasive monitoring system capable of simultaneously estimating blood glucose and cholesterol levels by integrating physiological multisensors with machine learning. The main contribution of this research is the development of a portable platform that combines Near-Infrared (NIR) sensors, photoplethysmography (PPG), and skin temperature sensing with a Random Forest regression model for simultaneous glucose and cholesterol prediction. Methods: The research method used is the experimental research and development (R&D) method with the ADDIE model (Analysis, Design, Development, Implementation and Evaluation). The system developed uses a combination of multi-wavelength near-infrared (NIR), photoplethysmography (PPG) and skin-temperature sensors and a Random Forest Regressor to estimate blood glucose and cholesterol levels. Sensor data were pre-processed by missing value handling, outlier removal using Interquartile Range (IQR) method and Min-Max normalisation. Hyper parameters of Random Forest were optimised using GridSearchCV and model performance was evaluated using R² , MAE and MSE. The optimized models were subsequently integrated into the non-invasive monitoring prototype and evaluated in terms of prediction error, signal stability, response time, and user perception. Result: In the experiment, the optimized Random Forest model achieved R² of 0.4515, MAE of 0.1386 and MSE of 0.0299 in glucose prediction, and R² of 0.4703, MAE of 0.0804 and MSE of 0.0150 in cholesterol prediction. The experimental results showed an average prediction error of ±8 mg/dL for glucose and ±10 mg/dL for cholesterol at system level. The standard deviation of the sensor signals was maintained at 0.03-0.05 and the average system response time was 2.8s. Moreover, the user evaluation has provided a usability score of 4.75 out of 5 which is a high level of usability of the system. The feature importance analysis of the optimized Random Forest model showed that age, nir680 and nir610 were the most important predictors for glucose estimation, while nir680, nir860 and nir610 were the most important predictors for cholesterol estimation. The results are consistent with the multisensor design of the proposed system, where PPG and NIR signals provide complementary physiological information for non-invasive estimation. The combination of physiological multisensor data with an optimized Random Forest regression model, supported by technical testing, fast system response, stable sensor signals, and high usability, looks very promising for the simultaneous estimation of glucose and cholesterol levels without touching the person and for the development of portable, real-time, IoT-enabled personalized healthcare systems. Novelty: The novelty of this research lies in the development of a portable IoT-based device that integrates physiological multisensors and machine learning for simultaneous and non-invasive glucose and cholesterol estimation with better interpretability.