TECHCOMB

Volume 2 : Issue 1

A New Adaptive Network Based Fuzzy Inference System for Time Series Forecasting

Authors : Erol E.,Cagdas H. A.,Ufuk Y.,Eren B.

image

Abstract:

Fuzzy inference systems have been used to solve a lot of real-world problems. Adaptive network fuzzy inference system (ANFIS) is one of the most important fuzzy inference systems. ANFIS was originally proposed for prediction and regression problems. When ANFIS have been used for time series forecasting, the inputs of ANFIS have been generally other simultaneous time series in the literature. However, it is a well-known that lagged variables should be used to obtain better forecasts in a time series forecasting process. Also, some advantages can be obtained if lagged variables are used as inputs for ANFIS. In this study, a new ANFIS is proposed by redesigning ANFIS approach for time series forecasting problem. Fuzzy c-means method is used for fuzzification in the suggested ANFIS. In addition, in the proposed approach, it is not necessary to utilize fuzzy numbers for input memberships. And, the parameters of output membership function are determined by using particle swarm optimization method. The proposed method is applied to two real-world time series. In order to compare the forecasting performance of the proposed approach, some other forecasting methods available in the literature are also applied to the time series. As a result of comparison, it is clearly observed that the most accurate forecasts are obtained when the proposed ANFIS is used.