BioScan: IoT Enabled Blood Group and Glucose Detection Using Machine Learning

Abstract

The increasing prevalence of diabetes mellitus and the growing demand for rapid physiological parameter assessment have accelerated research into non-invasive healthcare monitoring technologies. Conventional blood glucose measurement requires invasive blood sampling through finger-prick methods, while blood group identification relies on laboratory-based serological testing performed by trained medical personnel. These approaches can be uncomfortable, time-consuming, and unsuitable for continuous monitoring applications. Recent advances in optical biosensing, embedded systems, and machine learning have created new opportunities for estimating physiological parameters using non-invasive techniques. This research paper presents a non-invasive blood glucose and blood group detection system based on photoplethysmography (PPG) signal analysis and machine learning prediction. The proposed system employs a MAX30102 optical sensor integrated with an ESP32 WROOM-32 microcontroller to acquire dual-wavelength PPG signals using Red (660 nm) and Infrared (940 nm) light from a fingertip. The acquired signals are processed in real time and transformed into ten physiological features including signal mean values, pulsatile amplitudes, perfusion index, optical ratios, normalized signal parameters, heart rate, and signal quality metrics. These features are transmitted via WiFi to a Python Flask server where a trained Gradient Boosting Machine Learning model predicts blood glucose concentration in mg/dL and classifies blood group among eight categories (A+, A?, B+, B?, AB+, AB?, O+, O?). A synthetic dataset comprising 1200 PPG samples with realistic physiological correlations was generated for model training and evaluation. Comparative analysis of Ridge Regression, Random Forest, Support Vector Regression, and Gradient Boosting models identified Gradient Boosting as the most effective approach, achieving a Mean Absolute Error (MAE) of 14.64 mg/dL, Root Mean Square Error (RMSE) of 19.23 mg/dL, and R² score of 0.7897. Experimental results demonstrate successful finger detection, feature extraction, wireless communication, machine learning inference, and real-time OLED display of predicted blood glucose and blood group values. The developed system demonstrates the feasibility of non-invasive physiological parameter estimation using optical sensing and machine learning techniques; however, it is intended solely as a research prototype and not as a medically certified diagnostic device.

Introduction

It begins by highlighting the growing global burden of diabetes and the limitations of traditional invasive glucose monitoring methods, such as finger-prick tests, which are painful, inconvenient, and reduce patient compliance. It also notes that blood group testing is equally important in healthcare but traditionally requires laboratory-based, invasive procedures.

To address these challenges, the study proposes using Photoplethysmography (PPG) signals collected through a MAX30102 optical sensor integrated with an ESP32 microcontroller. PPG measures blood volume changes using red and infrared light and can capture physiological information related to cardiovascular activity and blood properties.

The system extracts multiple features from the PPG signals (such as AC/DC components, perfusion index, ratios, and heart rate) and sends them via WiFi to a Flask-based server, where a Gradient Boosting machine learning model performs prediction. The system estimates both blood glucose levels and blood group classification (A, B, AB, O types), and displays results on an OLED screen in real time.

The research uses machine learning methods like Random Forest, SVR, and Gradient Boosting, with an emphasis on real-time, low-cost, portable healthcare monitoring. It also highlights that while optical sensing and AI show promise, accurate non-invasive glucose and blood group prediction remains challenging and requires further clinical validation.

The literature review explains that:

• Existing non-invasive methods (NIRS, Raman, PPG, etc.) aim to replace invasive blood tests.
• PPG is widely used due to its simplicity, low cost, and ability to capture physiological signals.
• Machine learning helps identify hidden patterns in biosignals.
• However, most systems focus on a single parameter and lack real-time embedded integration.

The research gap is the absence of systems that simultaneously estimate both glucose levels and blood group using a single PPG-based, real-time embedded platform.

Conclusion

The proposed Non-Invasive Blood Glucose and Blood Group Detection System successfully demonstrates the application of optical biosensing and machine learning techniques for physiological parameter estimation without requiring invasive blood sampling procedures. The system integrates a MAX30102 dual-wavelength PPG sensor, ESP32 WROOM-32 microcontroller, Flask-based machine learning server, and OLED display to create a compact, low-cost, and real-time healthcare monitoring prototype. The developed system acquires Red and Infrared photoplethysmography signals from a user’s fingertip and processes these signals to extract ten important physiological features related to blood flow characteristics, optical absorption properties, and cardiovascular activity. These extracted features are analyzed using machine learning algorithms to estimate blood glucose concentration and predict blood group classification. Among the evaluated models, the Gradient Boosting Regressor achieved the best overall performance with a Mean Absolute Error (MAE) of 14.64 mg/dL, Root Mean Square Error (RMSE) of 19.23 mg/dL, and coefficient of determination (R²) of 0.7897. Experimental implementation confirmed successful finger detection, stable PPG signal acquisition, feature extraction, wireless communication with the Flask REST API server, and real-time OLED visualization of predicted results. The implementation of an offline fallback prediction model on the ESP32 further improves system reliability and operational continuity during network unavailability.The research highlights the growing potential of combining photoplethysmography and machine learning for future non-invasive healthcare monitoring systems. The proposed prototype offers advantages such as portability, user comfort, low hardware cost, and real-time analysis capability compared with conventional invasive diagnostic methods. However, the current system is based on synthetic training data and requires further clinical validation using real patient datasets. Factors such as motion artifacts, environmental noise, skin variations, and sensor positioning may also affect prediction accuracy. Therefore, the developed system should be considered solely as a research prototype and not as a medically certified diagnostic device. Future improvements may include advanced deep learning techniques, wearable integration, cloud-based monitoring, mobile application support, and large-scale clinical testing for enhanced accuracy and practical healthcare deployment.

References

[1] T.M.Selvakumari, “ Blood group detection using fiber optics,” in Armenian Journal of Physics, vol. 4, issue 3, 2011, Pages: 165-168 [2] Soumil Chugh and Jaskirat Kaur, “Non-Invasive Hemoglobin Monitoring Device,” IEEE ICCC ,2015 [3] Jens Kraitl and Ulrich Timm, “Non-invasive sensor for an in vivo hemoglobin measurement,” Hartmut Ewald Institute of General Electrical Engineering, University of Rostock, Germany ,Sensors IEEE,2011 [4] Gayathri T, Rekha M, Naizathul Akmha S and K.Nithyakalyani, “Non invasive blood group detection using light emitting diode,” in International Journal of Pure and Applied Mathematics , Chennai:vol 119 , 412018, pp: 565-574 [5] Gaurav S. Mehare, Chetan G, Pinjarkar Anup V, Tembhe, and Nitin S. Khachane, “A Non-invasive Way to Determine Blood Type Based on Image Processing,” in International Research Journal of Engineering and Technology (IRJET), vol 5, Issue 4, India:2018, pp:2040-2043 [6] RajashreeDoshi and AnaghaPanditrao, “Non-Invasive optical sensor for haemoglobin determination,” in International journal of engineering Research and Application(IJERA),2013:vol.3,Issue 2, pp:559-562 [7] U.Timm, D.McGrath, E.Lewis ,J.Kraitl and H.Ewald, “ Sensor system for Non-Invasive Optical Hemoglobin Determination,”2009, Pp:1975-1978 [8] Nishtha Nagar, Aesha Shah, Aditya Singh and Shreya Akotiya , “Blood Group detection and mobile monitoring system,” in International Conference on Innovative Trends in electronic communication and application, 2015, pp:20-24 [9] EdwardJayWang,WilliamLi, Dongawkins ,TerryGernsheimer, Colette Norby-slycord and Shwetak N.Patel, “ Hema App:Noninvasive Blood screening of haemoglobin using smartphine cameras,” 2016, Pp:593-604

Copyright

Copyright © 2026 Apurva Sanjay Deuskar, Tasmiya Akhtar Attar, Rutvik Vijay Patil , Shashank Sashibhushan Kumbhar , Pruthviraj Sardar Patil , Prof. Miss Mrunal Pawar. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.