Inference procedure based on LM-test in ANFIS for constructing time series model

Abstract

The aim of this study is to develop modeling procedure in Adaptive Neuro Fuzzy Inference System (ANFIS) for forecasting time series data. The focus of the development is selecting optimal ANFIS model by using the statistical inference based on Lagrange Multiplier (LM) test. To date, there are several methods for selecting optimal ANFIS model, but there is no research which applied LM-test procedure for selecting inputs, determining membership functions (clusters) and generating fuzzy rules, especially for forecasting time series data. Theoretical study related to the proposed procedure is supported by simulation study. The simulation datasets which generated based on Autoregressive Integrated Moving Average (ARIMA), ARIMA-Outlier and Seasonal ARIMA models are used for constructing ANFIS models and for evaluating the proposed algorithm. The performance of ANFIS models are evaluated by minimizing RMSE value.