Nonlinear time-series analysis with non-singleton fuzzy logic systems

We initiate an investigation of the use of nonsingleton fuzzy logic systems (NSFLSs) in forecasting of financial markets. The abilities of NSFLSs to approximate arbitrary functions, and to effectively deal with noise and uncertainty, are used to analyze several time series. First we show how to construct NSFLSs and train them using recursive least squares, or backpropagation. Then we use them to build predictive models of discrete and continuous chaotic time series corrupted by additive noise. Finally, we present an example of how NSFLSs can be used to produce predicted estimates of future values of commodities, and baseline our results with linear regression. Our NSFLS outperforms the linear regression results.