A multicriteria optimization approach for the stock market feature selection

Dragana Radojičić1, Nina Radojičić2 and Simeon Kredatus1

  1. TU Wien, Institute of Statistics and Mathematical Methods in Economics
    Wiedner Hauptstr. 8/E105-01-05 FAM, 1040 Vienna, Austria
    {gagaradojicic, simeon.kredatus}@gmail.com
  2. Faculty of Mathematics, University of Belgrade
    Studentski trg 16, 11000 Belgrade, Serbia
    nina@matf.bg.ac.rs

Abstract

This paper studies the informativeness of features extracted from a limit order book data, to classify market data vector into the label (buy/idle) by using the Long short-term memory (LSTM) network. New technical indicators based on the support/resistance zones are introduced to enrich the set of features. We evaluate whether the performance of the LSTM network model is improved when we select features with respect to the newly proposed methods. Moreover, we employ multicriteria optimization to perform adequate feature selection among the proposed approaches, with respect to precision, recall, and F β score. Seven variations of approaches to select features are proposed and the best is selected by incorporation of multicriteria optimization.

Key words

Limit order book, multicriteria optimization, time-series, feature selection, machine learning

Digital Object Identifier (DOI)

https://doi.org/doi.org/10.2298/CSIS200326044R

Publication information

Volume 18, Issue 3 (June 2021)
Year of Publication: 2021
ISSN: 2406-1018 (Online)
Publisher: ComSIS Consortium

Full text

DownloadAvailable in PDF
Portable Document Format

How to cite

Radojičić, D., Radojičić, N., Kredatus, S.: A multicriteria optimization approach for the stock market feature selection. Computer Science and Information Systems, Vol. 18, No. 3, 749–769. (2021), https://doi.org/doi.org/10.2298/CSIS200326044R