Prediction of Bitcoin Price Change using Neural Networks

Abstract

In recent years, Bitcoin is rising and become an attractive investment for traders. Unlike stocks or foreign exchange, Bitcoin price is fluctuated, mainly because of its 24-hours a day trading time without close time. To minimize the risk involved and maximize capital gain, traders and investors need a way to predict the Bitcoin price trend accurately. However, many previous works on cryptocurrency price prediction forecast short-term Bitcoin price, have low accuracy and have not been cross-validatedThis paper describes the baseline neural network models to predict the short-term and the long-term Bitcoin price change. Our baseline models are the Multilayer Perceptron (MLP) and the Recurrent Neural Networks (RNN) models. Data used are Bitcoin's blockchain from August 2010 until October 2017 with 2-days period and the total amount of 1300 data. The models generated are predicting both for short-term and long-term price change, from 2-days until 60-days.The result shows that long-term prediction has a better result than short-term prediction, with the best accuracy in Multilayer Perceptron when predicting the next 60-days price change and Recurrent Neural Networks when predicting the next 56-days price change. Multilayer Perceptron outperforms Recurrent Neural Networks with accuracy of 81.3 percent, precision 81 percent, and recall 94.7 percent.