Taylor-based optimized recursive extended exponential smoothed neural networks forecasting method

Abstract

The development of a Time series Forecasting System is a major concern for Artificial Intelligence researchers. Commonly, existing systems only assess temporal features and analyze the behavior of the data over time, thus, resulting in uncertain forecasting accuracy. Although many forecasting systems were proposed in the literature; they have not yet answered the attending question. Hence, to overcome this problematic, we propose an innovative method called Taylor-based Optimized Recursive Extended Exponential Smoothed Neural Networks Forecasting method, abbreviated as TOREESNN. Briefly explained, the proposed technique introduces three ideas to solve this issue: First, building an innovative framework for forecasting univariate time series based on Exponential Smoothed theory. Second, designing an Elman Classifier model for uncertainty prediction in order to correct the forecasted values. And finally hybrading the two recurrent systems in one framework to obtain the final results. Experimental results demonstrate that the proposed method has a high accuracy both in training and testing data in terms of Mean Squared Error (MSE) and outperforms the state-of-the-art Recurrent Neural Networks models on Mackey-Glass, Nonlinear Auto-Regressive Moving Average time series (NARMA), Lorenz, and Henon map datasets.