A Cryptocurrency Price Forecasting Model by Integrating Empirical Mode Decomposition and LSTM Neural Networks

Abstract

Cryptocurrencies, such as Bitcoin and Ethereum, are digital assets that use cryptographic techniques to enable secure and decentralized transactions over the internet. Cryptocurrency prices exhibit highly nonlinear and non-stationary behavior, influenced by a wide range of financial and nonfinancial factors, including market liquidity, regulatory developments, technological advancements, security incidents, and geopolitical events. The unpredictable nature of these price fluctuations underscores the need for robust predictive models to aid investors in making informed financial decisions. In this paper, we propose EMD-LSTM, a novel hybrid model that integrates empirical mode decomposition (EMD) and long short-term memory (LSTM) networks to enhance the accuracy of cryptocurrency price forecasting. EMD is utilized to decompose raw price signals into intrinsic mode functions (IMFs), which help in handling non-stationarity and extracting meaningful patterns. LSTM, with its capability to capture long-term dependencies, is then applied to the decomposed signals to learn relevant temporal features from high-frequency historical data. Our experimental results demonstrate that the EMD-LSTM model significantly outperforms traditional forecasting methods, achieving superior RMSE and MAE scores. These findings highlight the potential of EMD-LSTM as an effective tool for traders, investors, and researchers seeking reliable cryptocurrency price predictions in volatile market conditions.