A Fuzzy-AR Model to predict human body weights
Hideaki Tanii, Hiroshi Nakajima, Naoki Tsuchiya, Kei Kuramoto, Syoji Kobashi, Yutaka Hata · 2012
This paper proposes a body weight prediction method using Fuzzy-autoregressive (AR) model. New Fuzzy-AR model is formed by including fuzzy membership function which changes AR parameter in autoregressive (AR) model. We employed 452 volunteers, and collected their body weight time-series data during 730 days. We use body weight data from 1st to 365th day as learning data to determine the Fuzzy-AR models. After AR parameters are determined by Yule-Walker equation, we calculate the order, p, of the AR model for each volunteer based on Akaike's Information Criterion (AIC). In our experiment, we predicted body weight change for next p days for those subjects. In the Fuzzy-AR model, we make a fuzzy membership function based on the order of the AR model. As the result, the Fuzzy-AR model obtained higher correlation coefficient between predicted and truth values than the AR model on all volunteers. In addition, the Fuzzy-AR model obtained smaller mean absolute prediction error than the AR model.