Fuzzy Logic Approach - Computational Intelligence in Time Series Forecasting - page 186

Figure 4.6(c). Performance of the Gustafson-Kessel clustering-based fuzzy model with
evaluation data (top) and prediction error (bottom)

Table 4.2(d). Simulation results for nonlinear plant modelling

With training data With evaluation data

SSE(train) = 0.3973
MSE(train) = 0.0040

SSE(eval.) = 0.1215
MSE(eval.) = 0.0012

It is to be noted that the fuzzy model generated used only three Takagi-Sugeno
fuzzy rules and six antecedent fuzzy sets (for two inputs), which are much less than
that generated by the Wang-Mendel method or its modified approach.

4.8 Fuzzy Model as Nonlinear Forecasts Combiner

The need to combine forecasts of a time series has been well understood for a long
time. It has already been mentioned in Chapter 3 that not just any arbitrary
combination of forecasts is decisive in providing an improved forecast, but it is