I. Made Wirawan, Triyanna Widiyaningtyas, Muchammad Maulana Hasan
Bitcoin is currently the most widely used digital currency. The popularity of bitcoin continues to increase and become an asset of investment. To handle with the erratic bitcoin price changes, a prediction method is needed to help its users in predicting the price in the future. By utilizing a method that is able to recognize the pattern of change in the data time series in a certain period of time can be known bitcoin price for several days ahead with a high degree of accuracy. This research uses experimental methods. Data obtained from www.coingecko.com on May 1, 2013 to June 7, 2019. Preprocessing stage includes attribute removal, stationary test, and differencing. Determination of the model candidate using correlogram method. The predictions are done using the Autoregressive Integrated Moving Average (ARIMA) method, which is capable of generating high accuracy in short-term predictions. Evaluating the prediction results using Mean Absolute Percentage Error (MAPE). The results showed that ARIMA (4,1,4) models resulted in predictions with the smallest MAPE, 0.87 for the next one-day prediction and 5.98 for the next seven days. Thus the ARIMA (4,1,4) model is feasible to be used as a predictive method of Bitcoin for one to seven days ahead. © 2019 IEEE.
Electrical Engineering Department, State University of Malang, Malang, Indonesia
