Long-term prediction of blood pressure time series using multiple fuzzy functions

Long-term prediction of mean arterial blood pressure (MAP) time series can help clinicians to select a proper treatment based on their diagnosis. In this way, this paper firstly introduces a new prediction method for time series prediction based on fuzzy functions (FF) in multi-model mode and applies it for forecasting MAP time series as a new application. The proposed model consists of three steps. First step is to estimate the missing values in MAP time series by a linear interpolation method and to denoise it by using the empirical mode decomposition (EMD) procedure. Second step is to reconstruct the phase space. Last step is to apply a predictive model based on fuzzy functions (FFs). This model consists of two parts: 1) identifying the model structure by Gustafson-Kessel (GK) clustering and 2) estimating the output of each cluster by multivariate adaptive regression splines (MARS). Results show that, the proposed FF-based MARS model is more accurate than ANFIS and FF-based ANFIS, and its results are in the range of standard values. Beside, by using different strategies for long-term prediction, multiple FF-based MARS models has best result in comparison to recursive and multiple-recursive strategies.