Günlük Bitcoin Değerini Tahmin Etmek İçin İstatistiksel ve Makine Öğrenimi Algoritmalarının Karşılaştırılması

Abstract

Increasing fluctuations in pricing and having great profit potential, utilization in advanced machine learning technologies to make robust predictions of cryptocurrencies especially bitcoin have attracted great attention in recent years. In this study, various statistical techniques; Moving Average Analysis and Autoregressive Integrated Moving Average and machine learning (ML) techniques; Artificial Neural Network, Recurrent Neural Network (RNN) and Convolutional Neural Network have been conducted and compared to predict the future value of Bitcoin cryptocurrency price. They have been applied for the univariate time series analysis with a window size of 32. To prove the usefulness of ML algorithms, and to show that the results of RNN is a better, mean squared error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) indicators have been applied. The study revealed that recurrent neural network yields better results than other methods in predicting daily Bitcoin price in terms of MSE, MAE and MAPE metrics. Besides, Wilcoxon-Mann-Whitney nonparametric statistic test is applied to test the performance between ARIMA and machine learning algorithms.

Untitled

“…Caporale et al (2018) establish the existence of correlation amongst past and present values of the BTC price. Many studies (Basher & Sadorsky, 2022;Ye et al, 2022;Aygün & Günay Kabakçı, 2021;Chen et al, 2020;Munim et al, 2019;Adcock & Gradojevic, 2019;Mallqui & Fernandes, 2019;Rizwan et al, 2019;McNally et al, 2018) confirm the robustness of the univariate approach. Ye et al (2022) apply an ensemble machine learning model to forecast Bitcoin's next prices.…”

Bitcoin Price Prediction with Fuzzy Logic

“…Figure 1 shows the daily closing price of 2500 days of Bitcoin. However, some researchers use the average between the highest and the lowest price, or even all four values at once [14]. We studied some of these indicators before deciding which one to use.…”

“…Two statistical and three machine learning algorithms are used to forecast the daily price of Bitcoin [14] in a comparison of statistical and machine learning techniques. The simplest method, moving average analysis, which forecasts future values by averaging previous n values, has the biggest inaccuracy.…”