Improvement of LSTM-Based Forecasting with NARX Model through Use of an Evolutionary Algorithm

The reported work aims to improve the performance of LSTM-based (Long Short-Term Memory) forecasting algorithms in cases NARX (Nonlinear Autoregressive with eXogenous input) models by using evolutionary search. proposed approach, ES-LSTM, combines a two-membered ES local search procedure (2MES) an ADAM optimizer train more accurate LSTMs. accuracy is measured from both error and trend prediction points view. method first computes learnable parameters LSTM, subset training data, applies modified version 2MES optimization tune them. In second stage, all available data are used update LSTM’s weight parameters. resulting algorithm assessed versus standard trained LSTM case multiple financial time series. tests conducted on test respectively. experimental results show significant improvement direction change without damaging measurements. All quality measures better than algorithm, while insignificantly higher or, some cases, even better. Together theoretical consideration, this proves that new outperforms one.