FUZZY LOGIC-BASED LONG SHORT-TERM MEMORY NEURAL NETWORK ARCHITECTURES FOR FINANCIAL TIME SERIES FORECASTING

Read full text Summarise with VIRAT
Authors: A.G. Makhova M.I. Kumskov
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Forecasting financial time series remains a challenging task due to high volatility, nonlinear relationships, and the presence of noise in market data. Traditional statistical approaches often struggle to capture these complex patterns, while deep learning methods have demonstrated promising results in modelling sequential dependencies. In this study, we investigate a neural network architecture that combines Long Short-Term Memory (LSTM) networks with principles of fuzzy logic for financial time series forecasting. The proposed approach introduces fuzzy logic both at the data preprocessing stage and within the neural network itself. During preprocessing, fuzzification of input data is used to represent financial observations as fuzzy sets, allowing the model to better handle uncertainty and reduce sensitivity to noise. In addition, fuzzy concepts are incorporated into the network through modified activation functions, which can be interpreted as membership functions of fuzzy sets. This design enables the model to perform operations analogous to fuzzy inference while learning temporal dependencies in the data. Experimental results on financial datasets demonstrate that the proposed fuzzy-enhanced LSTM architecture improves forecasting accuracy compared with a standard LSTM model. The findings suggest that integrating fuzzy logic with recurrent neural networks provides a promising direction for improving financial time series prediction.

Fields of Study & Subjects

Stock Market Forecasting Methods Forecasting Techniques and Applications Hydrological Forecasting Using AI

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
25 March 2026
Language
EN