Predicting Bitcoin Price: Comparative Analysis of Machine Learning and Deep Learning Models

Abstract

In recent years, Bitcoin has become an important financial instrument that has attracted increasing attention as a digital currency. Accurately predicting the value of a financial asset is of great importance for both individual and institutional investors. The aim of this study is to evaluate and compare the predictive power of different models (Support Vector Regression (SVR), Convolutional Neural Network (CNN), Long Short- Term Memory (LSTM), Hybrid model, which is a combination of CNN and Bidirectional LSTM (CNN-BiLSTM), and XGBoost) in predicting the Bitcoin price. The main objective of the study is to determine the most effective algorithm in predicting the Bitcoin price. In the study, external factors such as S&P500 index, Gold/Dollar exchange rate, West Texas Intermediate Oil Price and Dollar Index were used to predict the Bitcoin price. The dataset covers 2191 days of data between January 1, 2015 and September 18, 2023. The models went through a two- stage process consisting of training and testing stages. The performance of the models is evaluated using various statistical metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE) and R-squared (R2). The results show that the XGBoost algorithm gives the best results in all performance metrics. The XGBoost model is followed by CNN-BiLSTM, CNN and LSTM models, respectively. The SVR model exhibited the lowest performance.