Health Monitoring with Smartphone Sensors and Machine Learning Techniques
Rahul Kusuma, Shyamapada Mukheerjee · 2023
Over the past few decades, there has been a noticeable increase in life expectancy globally. However, this increase is anticipated to result in a significant rise in the aging population, in conjunction with a declining birth rate. Conventional approaches for estimating calorie expenditure necessitate the usage of wearable devices, such as smartwatches, pedometers, or smart bracelets, that track user activity continuously and estimate the amount of energy expended. In this work, accelerometer and gyroscope data from smartphones are utilized to provide a way to estimate calories burned while performing certain activities. Human activity recognition (HAR) techniques are employed to classify activities performed based on accelerometer and gyroscope data, and the resulting activity labels are used along with the MET (Metabolic Equivalent of Task) score of these activities and BMI (Body Mass Index) to estimate calories burned during a particular activity. HUman Activity Recognition models achieve an accuracy of about 94% with random forest and 97% with Convolutional Neural Networks.