# Forecasting the movements of Bitcoin prices: an application of machine learning algorithms

**Authors:** Hakan Pabuccu, Serdar Ongan, Ayse Ongan  
**arXiv:** [2303.04642](https://arxiv.org/abs/2303.04642)  
**Date:** 2023-03-08  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.04642v1/

## Key points

- Cryptocurrencies, particularly Bitcoin, are controversial and complex technological innovations in today's financial system
- Forecasting the movements of Bitcoin prices with a high degree of accuracy is important research
- Four different Machine Learning (ML) algorithms were applied simultaneously: Support Vector Machines (SVM), Artificial Neural Network (ANN), Naive Bayes (NB), and Random Forest (RF)
- Logistic regression (LR) was used as a benchmark model for comparison purposes
- Nine technical input parameters were used, including closing prices, high and low prices for computing technical indicators such as simple 14-day moving average, simple 14-day weighted moving average, momentum, stochastic K%, stochastic D%, relative strength index(RSI)
- The study applies four different ML algorithms simultaneously to compare their performances and uses both existing continuous datasets and discrete datasets created specifically for this study
- The study aims to understand which ML algorithm offers higher forecasting performance in continuous and discrete datasets separately and comparatively
- RF has the highest forecasting performance in the continuous dataset while NB has the lowest performance. On the other hand, ANN has the highest performance while NB has the lowest in the discrete dataset.
- Using a discrete dataset improves overall forecasting performance across all models estimated.
- A validation dataset consisting of 100 Bitcoin series between June 2020–October 2020 was used to compare model performances with benchmark LR statistics using t-tests.
- Accuracy statistics were calculated for both continuous and discrete datasets.
- This study provides valuable insights into the forecasting of Bitcoin prices using ML algorithms.

## Summary

Cryptocurrencies, particularly Bitcoin, have emerged as one of the most controversial and complex technological innovations in today's financial system. The ability to forecast the movements of Bitcoin prices with a high degree of accuracy has become an important area of research. This study aims to achieve this objective by applying four different Machine Learning (ML) algorithms simultaneously, namely, Support Vector Machines (SVM), Artificial Neural Network (ANN), Naive Bayes (NB), and Random Forest (RF). Additionally, logistic regression (LR) is used as a benchmark model for comparison purposes. The study uses nine technical input parameters that were previously identified by Armano et al. (2005), Atsalakis & Valavanis (2009), Kara et al. (2011), and Kim (2003). These parameters include closing prices, high and low prices for computing technical indicators such as simple 14-day moving average, simple 14-day weighted moving average, momentum, stochastic K%, stochastic D%, relative strength index(RSI). The study differs from previous studies in three aspects. Firstly, it applies four different ML algorithms simultaneously to compare their performances. Secondly, it uses both existing continuous datasets and discrete datasets created specifically for this study. Finally, the study aims to understand which ML algorithm offers higher forecasting performance in continuous and discrete datasets separately and comparatively. The empirical findings reveal that RF has the highest forecasting performance in the continuous dataset while NB has the lowest performance. On the other hand, ANN has the highest performance while NB has the lowest in the discrete dataset. Furthermore, using a discrete dataset improves overall forecasting performance across all models estimated. To validate these findings further, a validation dataset consisting of 100 Bitcoin series between June 2020–October 2020 was used to compare model performances with benchmark LR statistics using t-tests. Accuracy statistics were calculated for both continuous and discrete datasets. In conclusion, this study provides valuable insights into the forecasting of Bitcoin prices using ML algorithms. The use of both continuous and discrete datasets along with the application of multiple ML algorithms contributes to a better understanding of which algorithm offers higher forecasting performance.