Implementing ANFIS with Three-Layer Architecture for Time Series

The complex time addiction and randomness of multivariate time series make it nec-essary to apply time series analysis to these variables. So, it is important to produce methods that can be used appropriately for time series system identification. Time series are generally handled as a single-layer architecture consisting of only the observed data processing layer. In this study, a hybrid method has three-layer adapted and evaluated using an adaptive neuro-fuzzy inference system (ANFIS). The main motivation of this study is to learn the errors produced by ANFIS method and to use them as new in-formation. For this purpose, proposed method was evaluated using real-time serial data sets obtained from different fields. Due to the three-layer architecture, the errors caused by the results produced by ANFIS are reused. As a result, the method of learning from errors has been realized and better results have been produced com-pared to traditional single-layer architecture.