Machine Learning Based Fitness Tracker Platform Using MEMS Accelerometer

Abstract

This paper deals with real time development of a machine learning based portable system for determining eating habits of a human being using six point calibrated wearable MEMS tri-axial accelerometer. The rise of obesity as a global epidemic makes it immensely important to monitor food habits of a modern day person. In this proposed system, we have derived an easy to adopt algorithm based on a training based model for identifying the amount of calories consumed and burnt by a person. The proposed system consists of a wrist worn MEMS accelerometer that calculates calories burned per step which is directly sent over to the user's smart phone and a cloud based machine learning algorithm that does the prediction of health habit (i.e. healthy, unhealthy or undernutrition) based on the data obtained from the wrist worn device. In order to calculate the health habit of the user, the cloud uses logistic regression with calories burnt (from MEMS accelerometer) and calories consumed (daily manual input)to predict health habit of the user. The wrist worn device extracts calories burnt per step from the change in Y-axis acceleration data of the accelerometer in the wearable device, which after self-calibration is sent over to the user's smart phone through Wi-Fi. Thus, this cloud based food habit detection not only decreases the risk of obesity in a person but also introduces a low cost alternative device with reduced power consumption of $(< 13.5 \text{mW})$ and minimal covering size (12.56cm2) that can improve people's life.