IoT-Based Deep Neural Network Approach for Heart Rate and SpO2 Prediction

A deep neural network (DNN) is the most effective machine learning tool to improve the errors of the biomedical instrumentation system. The main purpose of this work is to design and implement a DNN-based system to predict the heart rate (HR) and SpO₂ values. In the proposed work, the sensor is used to detect the HR and SpO₂ of humans. The performance of the proposed work without using DNN has Root Mean Square Error (RMSE) of 1.14 and 0.84, Mean Absolute Relative Difference (MARD) of 1.1 and 0.7, Coefficient of Determination R² = 97% and R² = 30%, and accuracy of 88.55% and 92.73%, for HR and SpO₂, respectively. The DNN model is proposed to improve performance. The sensor output is given to the PCA algorithm, and then these values are used to train and validate the DNN model. The performance of the system is improved remarkably using the DNN model and achieved the RMSE of 0.32 and 0.35, MARD of 0.1 and 0.1, R² = 99% and R² = 90%, and accuracy of 98.72% and 98.98%, for HR and SpO₂, respectively. The cloud-based system is implemented on the Raspberry pi-4 using a python programming language. The proposed work has some additional features, such as it sends all parameters to the cloud for further processing or future use and the measured parameters to WhatsApp and SMS via the cloud. Alert messages will be sent to the concerned person if the HR and SpO₂ levels cross the critical level.