The impact of stock market price Fourier transform analysis on the Gated Recurrent Unit classifier model

Abstract

In this paper, we suggest new feature extraction models based on the stock market price signal analysis. In particular, we study the behavior observed in signals originating from different sources, such as prices of different Limit Order Book levels and open, close, low, high prices of the preselected time intervals. We apply Fourier transformation to extract new features. Moreover, we evaluate if the performance of the model based on the Gated Recurrent Unit (GRU) architecture is improved when we select features utilizing the proposed methods. Furthermore, we benchmark the performance of new indicators on the GRU model and provide quantified results proving the significant performance improvement obtained by incorporating the suggested features.