Parameter selection for long short-term memory networks with multi-criteria decision-making tools: an application for G7 countries stock market forecasting

This study integrates Long Short-Term Memory (LSTM) networks with Multi-Criteria Decision-Making (MCDM) methods to improve the accuracy of stock market forecasts. Drawing on a dataset from G7 stock markets spanning June 2018 to June 2023, the study aggregates fifteen performance metrics to generate a diverse parameter pool with randomly assigned values. These parameters are evaluated using decision matrices applied through the MARCOS (Measurement of Alternatives and Ranking according to Compromise Solution) and TOPSIS (Technique for Order of Preference by Similarity to Ideal Solution) methods. The models are assessed under both equal-weighted criteria and criteria weighted via the CRITIC (Criteria Importance through Intercriteria Correlation) method. Additionally, the study examines Empirical Mode Decomposition-based LSTM (EMD-LSTM) models, revealing that those optimized using the MARCOS method substantially outperform those optimized through TOPSIS, particularly in forecasting accuracy. The integration of MCDM techniques with LSTM models yields a hit rate of up to 75%, demonstrating the effectiveness of this approach in parameter selection and overall model enhancement. This study not only underscores the potential of MCDM methods in refining LSTM models but also provides a robust framework for improving predictive accuracy in financial markets.