BMI Estimation from 2D Face Images Using Support Vector Machine
Joshua C. Gonzales, Joshua Ron G. Garcia, Jocelyn F. Villaverde · 2022
Several studies on predicting Body Mass Index (BMI) based on face images have already been conducted. However, a lack of study employs machine learning to predict BMI. Furthermore, these studies only include Caucasian and African subjects, indicating that Asians are lacking. A system that utilizes the Support Vector Machine (SVM) algorithm was developed in this study to classify BMI from 2D face images by analyzing facial features. The Raspberry Pi 3 Model B+ was employed as the main microcomputer, and Raspberry Pi Camera version 1.3 was utilized to capture face images. The classifier model was trained with five hundred two (502) samples, including various ethnicities, age groups, and weight categories. They are underweight, normal, overweight, and obese. The system was tested on twenty-five (25) subjects for each of the four (4) BMI classes with a total of one hundred (100) samples. The results of the system were evaluated using a confusion matrix and obtained an accuracy score of 91%. This demonstrates that the SVM for BMI estimation has a high accuracy rate.