Predicting Bitcoin and Ethereum Prices Using the Long Short- Term Memory (LSTM) Model

Abstract

Cryptocurrency is a highly volatile digital asset, necessitating accurate and adaptive forecasting methods. This study implements a Long Short-Term Memory (LSTM) model to predict the daily closing prices of two leading cryptocurrencies Bitcoin (BTC) and Ethereum (ETH) using historical data from Yahoo Finance and Binance. To enhance data richness and model robustness, datasets from both sources were vertically merged. The methodological framework included data preprocessing, Min–Max normalization, formation of 24-day sliding input windows, and training across three data split ratios (70:30, 80:20, and 90:10). Model performance was evaluated using the Root Mean Squared Error (RMSE). Results indicate that the LSTM model achieved high prediction accuracy, with the lowest RMSE values of 0.0137 for BTC and 0.0152 for ETH using the combined dataset with a 90:10 split. Beyond modeling, a web-based application was developed using Streamlit, enabling users to perform real-time predictions and export results. This study contributes to the field of cryptocurrency forecasting by demonstrating that multi-source data integration significantly improves predictive accuracy and model generalization. The proposed framework offers both theoretical insights and practical tools for researchers and investors in financial technology.