Forecasting Bitcoin Closing Price by Four Machine Learning Algorithms

Abstract

Bitcoin has increased in popularity as a speculative asset. Since 2013, eventually becoming the most recognizable cryptocurrency. But it's worth noting that the price of Bitcoin has a very high degree of volatility and diversity, which means the ability to estimate prices accurately is crucial for making wise financial decisions. Although recent research has implemented machine learning to predict Bitcoin prices with greater precision, such as Long short-term memory (LSTM), few have focused on traditional machine learning methods. In this article, the author chose a data set including nearly eight years of daily bitcoin price data for closing price prediction. Four different machine learning algorithms were used simultaneously: the Linear Regression (LR), the Decision Tree (DT) and the Random Forest (RF). An artificial neural network, the Multilayer Perceptron (MLP) was also used in this study. The author altered parameter values using the cross-validation method before creating the models in order to get more precise predictions. Finally, Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and R-squared are used as indicators to assess the outcomes from each model. The study's findings demonstrated that all three metrics of Linear Regression outperformed the performance of the other three models. Perhaps future research could focus more on traditional machine learning algorithms instead of going after complex models.