Title: 運用機器學習方法分析比特幣交易行為與貨幣匯率之交互關係
A machine learning approach to analyze the relationship between Bitcoin transaction behavior and currency exchange rates
Authors: 劉銘騏
Liu, Ming-Chi
Chen, An-Pin
Huang, Szu-Hao
Keywords: 比特幣;自適應增強;類神經網路;匯率;支援向量迴歸;Bitcoin;Adaboost;Exchange rate forecasting;Affinity Propagation;Neural Network;Support Vector Machine
Issue Date: 2017
Abstract: 身為最大的數位貨幣,比特幣有著許多傳統貨幣沒有的特色,包含了極高的波動度,去中心化的管理方式以及區塊鏈分類帳本,其中在區塊鏈這部分,記錄著從比特幣誕生到今日的每一筆交易,這使得比特幣比一般的現實貨幣擁有了更多的資訊,可以藉由分析這些交易者的紀錄來得知一些隱藏的知識。本研究中測試了數種貨幣特徵、六個交易者特徵以及三個機器學習方法,期望能找出比特幣匯率變動最精準的預測及建模方式。而最後結果顯示,現實貨幣匯率對比特幣匯率的影響微乎其微,比特幣交易人的行為對比特幣匯率的影響則非常的大,而最適合用來預測比特幣匯率的模型是Adaboost。
As the most popular digital currency, Bitcoin has its own unique characteristics, such as decentralize management and blockchain ledger technology. In addition, its volatility is much greater than traditional currencies. The blockchain records every transition since the beginning of Bitcoin. From the perspective of big data analytics, Bitcoin contains rich and valuable transition information, which may help the discovery of hidden knowledge about transistors. Our research is expected to precisely predict the exchange rate of Bitcoin through six different features from transistors, three different machine learning algorithms, and various kinds of currency attributes. The experimental results show that the proposed system can effectively model the relationship between the exchange rate change and the transition behavior of Bitcoin. It also displays that traditional currencies seldom influence the change rate of Bitcoin, while Bitcoin users’ behavior is an important factor that makes tremendous difference when forecasting Bitcoin change rate. Finally, compared to artificial neural networks and support vector machines, the learning model based on Adaboost algorithm can achieve the most accurate prediction results.
URI: http://etd.lib.nctu.edu.tw/cdrfb3/record/nctu/#GT070453424
Appears in Collections:Thesis