High-accuracy low-precision machine learning system for health monitoring

• Main Author: Ang, Wei Jun
• Other Authors: Mohamed M. Sabry Aly
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/162809
• Institution: Nanyang Technological University

Background The increasing rates of cardiovascular diseases (CVDs) within global and local populations is alarming. Although older adults have higher risks of developing CVDs (Rodgers et al., 2019), it is crucial for individuals of all ages to be mindful of their cardiovascular health and take preventive measures or initiate medical care if necessary. This could be done through cardiovascular health monitoring, which allows detection of early symptoms of CVD; a common one being abnormality in heart rate. A wide array of devices for cardiovascular health monitoring purposes have been developed in recent years, ranging from wireless electrocardiogram (ECG) monitors to wearable gadgets such as smartwatches. Given the broad spectrum of heart rates of individuals of different age groups and backgrounds, the accuracy of such devices in detecting heart rate abnormality is a pivotal aspect in the development of these devices. Objective The objective of this project is to develop a high accuracy, low-precision machine learning system to alert users when abnormalities in heart rates for various activities are detected. Methods A Raspberry Pi (RPI) was used as an intermediary for the Himax and smart sensor watch to communicate. Upon booting up the RPI, the smart sensor watch transmits real-time heart rate data from the user to the RPI, which was forwarded to the Himax. Accelerometer data together with the heart rate data was fed into the model for inference. In accordance with the intensity of activity conducted, when an abnormality in heart rate is detected, users will be alerted through a red light on the LED. Results The machine learning system is able to classify correctly with a rate of 99.46% Recommendation Functions such as emergency response and Bluetooth communication between the health sensor band and RPI can be implemented in the future with sufficient time and resources.