A New evolving Fuzzy System with Mechanisms to Deal With Uncertainties in Times Series Forecasting

In the information era, many of the collected data take the stream form, which is usually non-stationary and produced at high speed. Furthermore, the real-world data sets inevitably contain unnecessary and irrelevant information called noise, disturbing the model’s accuracy. This paper proposes recursive filtering to smooth the input data. This approach is implemented in the so-called evolving Participatory Learning with Kernel Recursive Least Square and Distance Correlation (ePL-KRLS-DISCO). The such implementation makes the model more robust when dealing with noise data, reducing the model’s error. The introduced model is tested using the Mackey-Glass time series and the nonlinear system identification with the injection of noise. The performance is evaluated in terms of error metrics and the number of final rules. Furthermore, the results are compared with the results of classical models and evolving Fuzzy Systems. Among the six simulations, the proposed model outperformed the compared models in five concerning the errors.