Estimation of Anthropometric Measurements Using Optimized Machine Learning Models with Bayesian Algorithm

Document Type : Article


1 Bayburt Üniversitesi

2 Bursa Teknik Üniversitesi

3 Meram


This study collects the anthropometric measurements and weights of 185 male individuals between 55 and 65 years old from Ankara city of Turkey. A total of 29 variables with three inputs and twenty-six outputs are collected. This paper aims to develop machine learning-based models to estimate anthropometric measurements from weight, stature, and eye height. These models are support vector regression (SVR) optimized with Bayesian based on quadratic kernel, Gaussian Process Regression (GPR) optimized with Bayesian based on matern5/2 kernel. This study contributes to SVR and GPR models by using Bayesian method to optimize the parameters as a difference from the literature. According to the literature review, applying these two models to anthropometric measurements for the first time is predicted. The estimation results are compared based on three metrics, namely Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE). GPR optimized with Bayesian model has better accuracy than SVR optimized with Bayesian for all combinations except interpupillary distance, according to the obtained results. The RMSE values of the best models selected for each combination varied between 0.255 and 0.319 during the testing phase. Especially the estimations made with GPR optimized with Bayesian have a shallow error rate.


Main Subjects