Bitcoin Price Prediction Using Deep Bayesian LSTM With Uncertainty Quantification: A Monte Carlo Dropout–Based Approach

Abstract

Bitcoin, being one of the most triumphant cryptocurrencies, is gaining increasing popularity online and is being used in a variety of transactions. Recently, research on Bitcoin price predictions is receiving more attention, and researchers have investigated the various state‐of‐the‐art machine learning (ML) and deep learning (DL) models to predict Bitcoin price. However, despite these models providing promising predictions, they consistently exhibit uncertainty, which cannot be adequately quantified by classical ML models alone. Motivated by the enormous success of applying Bayesian approaches in several disciplines of ML and DL, this study aims to use Bayesian methods alongside Long Short‐Term Memory (LSTM) to predict the closing Bitcoin price and consequently measure the uncertainty of the prediction model. Specifically, we adopted the Monte Carlo dropout (MC‐Dropout) method with the Bayesian LSTM model to quantify the epistemic uncertainty of the model's predictions and provided confidence intervals for the predicted outputs. Experimental results showed that the proposed model is efficient and outperforms other state‐of‐the‐art models in terms of root mean square error (RMSE), mean absolute error (MAE) and R2. Thus, we believe that these models may assist the investors and traders in making critical decisions based on short‐term predictions of Bitcoin price. This study illustrates the potential benefits of utilizing Bayesian DL approaches in time series analysis to improve data prediction accuracy and reliability.