ABSTRACT
Body Mass Index (BMI) is an important and useful indicator for medical diagnoses, accurate monitoring and forecasting of BMI are therefore crucial. However, the current measurement of BMI, which is usually highly correlated with the environmental and individual conditions, is inaccurate. Recent developments of bioelectrical impedance show that there is a great potential to improve the measurement of BMI. In this paper, we propose a novel interpretable Takagi-Sugeno Fuzzy NARX (TSF-NARX) model to predict BMI values from bioimpedance signals and anthropometric factors. The proposed model integrates the Nonlinear Auto Regressive Moving Average with Exogenous Input (NARMAX) method and Takagi-Sugeno fuzzy inference. An obvious novelty and advantage of the proposed method is that it provides a new framework, combining the capabilities of fuzzy inference and NARX representation empowered by nonlinear membership functions. The experimental results show that the TSF-NARX model outperforms other models in prediction accuracy and consistency. More importantly, the model identifies both the key frequency bands and anthropometric factors that highly affect the BMI. The proposed model provides a tool for obtaining accurate, interpretable and robust measurement against the intra and extra uncertainty within the clinical diagnosis.
