Design of an IoT-Based Mass Index (BMI) Prediction Mode

Abstract

Overweight and obesity have a significant economic and social implications in terms of low productivity, high mortality rate and increased health care needs. Overweight and obesity have become a major health concern associated with diseases such as cardiac arrest, type 2 diabetes, stroke, high blood pressure, and other non-communicable diseases (NCD) and are the leading risks for deaths globally, killing more people than underweight. Body Mass Index (BMI) is a measure that uses weight and height to work out a person’s nutrition status. Research throughout to calculate BMI is based on traditional manual methods which are time consuming, error prone and they are not cloud-based. Few systems have incorporated machine learning yet with low accuracy. Existing literature is based on areas with high number of overweight and obese cases, however, lacking information from regions in transition. Based on these findings this research takes a technological approach of calculating BMI among the residents of Kitengela town Kajiado County in Kenya men and women (aged between 5 and 50 years) using an IoT based BMI system. This system consists of a NodeMCU microcontroller for computations with an inbuilt ESP8266 WiFi module for connectivity to the internet, load cell sensor for body weight measurement, a HX711 load cell amplifier module and HC-SR04 ultrasonic sensor for height measurement. Values are displayed on a 16x2 LCD and sent to ThingSpeak for storage and analysis. ThingSpeak is integrated with MATLAB Machine Learning to make the prediction based on height and weight sensory data. This research uses Supervised Exponential Gaussian Process Regression algorithm to predict whether a person is underweighted, normal weight, overweight or obese. The designed IoT Based BMI computation system achieves an accuracy of 99.18% with a time reduction of 1.1 % per person while the ML model achieves an accuracy of 98%. The system was more time efficient in that the 45 residents were measured in 15minutes while the manual system took 4 hours 5 minutes, a time saving factor of 3 hours 50 minutes.