FuzzyLSTM and Financial Time Series Representations
Received: 10 Oct 2026 Revised: 20 Oct 2026 Accepted: 10 Nov 2026
Published: 2026, vol. 30, issue 4, pp. 365–368
Abstract
The effect of financial time series representation on forecast quality is investigated. FuzzyLSTM is considered with an additional feature computed from price history, parametric sigmoid functions in the LSTM gates, and blending of the network forecast with the latest observed price. Sufficient conditions for the ISS stability of the recurrent component are formulated, and assumptions ensuring boundedness of the final forecast are specified. On minute, hourly, and daily BTC–USD data, the ratio of the standard LSTM RMSE to the FuzzyLSTM RMSE was \(1.24\), \(1.45\), and \(1.54\), respectively. Adding the selected technical indicators did not provide a systematic improvement in FuzzyLSTM, while, in a separate comparison, symbolic Transformer input showed no consistent advantage over numerical input.
Keywords: financial time series, forecasting, FuzzyLSTM, technical indicators, symbolic representation.
BibTeX
@article{IS-Kumskov-Makhova-Medvedev-Zhuravlev2026,
author = {Kumskov, Mikhail Ivanovich and Makhova, Anastasia Gennadievna and Medvedev, Petr Alekseevich and Zhuravlev, Georgy Valerievich},
title = {{FuzzyLSTM and Financial Time Series Representations}},
journal = {Intelligent Systems. Theory and Applications},
year = {2026},
volume = {30},
number = {4},
pages = {365--368},
}
AMSBIB
\Bibitem{IS-Kumskov-Makhova-Medvedev-Zhuravlev2026}
\by M.\,I.~Kumskov, A.\,G.~Makhova, P.\,A.~Medvedev, G.\,V.~Zhuravlev
\paper FuzzyLSTM and Financial Time Series Representations
\jour Intelligent Systems. Theory and Applications
\yr 2026
\vol 30
\issue 4
\pages 365--368
\lang In Russian
RU