An Intelligent System for Obesity Detection through Human Activity Recognition and BMI Prediction
Khorshed Alam, Diganta Chowdhury, Promila Haque · 2023
Obesity is a global epidemic affecting people of all ages and genders. An imbalance between the calories consumed and burned in energy is the primary cause of this disease. To combat this issue, researchers have developed an intelligent system based on supervised and unsupervised techniques to detect obesity levels and assist consumers and healthcare professionals in leading healthier lifestyles. Our proposed method uses a public domain dataset, WISDM human activity recognition, and a 3-layered LSTM to recognize physical activities accurately with 99% accuracy. The technology involved in recognizing human’s physical movement is also known as Human Activity Recognition. Using a smartphone's built-in sensors (accelerometer), daily activities can be identified, and the total number of calories expended can be calculated, along with calorie intake based on food habits. Users can set goals to track their physical activity and forecast the timeframe towards obesity for various periods. The deep learning model is deployed as .pb file in Android Studio where it uses the accelerometer sensor of a smartphone to detect human body movement in real-time for a specific timeline, and the user's BMI can be predicted for the upcoming one month, six month, and one-year periods. With this information, the likelihood of developing obesity within at least a year can be predicted. Overall, this system can assist users in leading healthier lifestyles and combatting the global epidemic of obesity. We test our system with 15 participants detecting obesity for next one year (2022-2023). By evaluating results, we can see how robust and accurate our system is.