Comprehensive Comparison of LSTM Variations for the Prediction of Chaotic Time Series

Abstract

Chaos theory is a field of study used to model the irregular fluctuations encountered in all areas of our daily life. Time series obtained from chaotic systems such as currencies, stock markets, and weather conditions are called chaotic time series. Making predictions from these time series has recently attracted the attention of researchers as it will eliminate the need for complex mathematical models. Recently, deep neural networks are widely used in the prediction of chaotic time series as well as in solving many problems in the literature. In this study, time series obtained from three different chaotic systems were collected in order to predict chaotic time series, and four different LSTM variations were used to predict these data. As a result of the experimental studies, it has been observed that Stacked LSTM predicts more successfully than other LSTM variations. The obtained 3.7397x10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-5</sup> validation RMSE value and 0.1558 test RMSE value show that Stacked LSTM gives more accurate results than other methods used in the prediction of chaotic time series in the literature.