Time series prediction using crisp and fuzzy neural networks: a comparative study

Abstract

Every organization needs adequate forecasts for planning the future. The accuracy of forecasts is influenced by both the quality of past data and the method selected to forecast the future. In this paper, we carry out a comparative study between the time series forecasts from (1) the Quick-prop neural network, (2) a fuzzy neural network (adaptive-network-based fuzzy inference system (ANFIS)), (3) a fuzzy regression and identification decision tree (ADRI), and (4) traditional time series methods (ARIMA models). We augment ANFIS by using fuzzy curves to identify the input variables that have the most influence on the output. This method identifies the significant input variables that lead to a considerable decrease in training time for ANFIS, while keeping the performance at least as good. We test the performance of ANFIS with the fuzzy curve pruning technique on empirical time series data (the national private consumption) from the Spanish economy. ANFIS produced the best performance on forecasting the empirical time series data compared to ADRI and ARIMA.