Proceedings of the 1st International Conference on Statistics and Analytics, ICSA 2019, 2-3 August 2019, Bogor, Indonesia

Research Article

An Empirical Study in Forecasting Bitcoin Price Using Bayesian Regularization Neural Network

Download25 downloads
  • @INPROCEEDINGS{10.4108/eai.2-8-2019.2290515,
        author={Rina  Sriwiji and Arum Handini  Primandari},
        title={An Empirical Study in Forecasting Bitcoin Price Using Bayesian Regularization Neural Network},
        proceedings={Proceedings of the 1st International Conference on Statistics and Analytics, ICSA 2019, 2-3 August 2019, Bogor, Indonesia},
        publisher={EAI},
        proceedings_a={ICSA},
        year={2020},
        month={1},
        keywords={bayesian regularization neural network bitcoin regression},
        doi={10.4108/eai.2-8-2019.2290515}
    }
    
  • Rina Sriwiji
    Arum Handini Primandari
    Year: 2020
    An Empirical Study in Forecasting Bitcoin Price Using Bayesian Regularization Neural Network
    ICSA
    EAI
    DOI: 10.4108/eai.2-8-2019.2290515
Rina Sriwiji1, Arum Handini Primandari1,*
  • 1: Department of Statistics, Universitas Islam Indonesia, Yogyakarta, 55584, Indonesia
*Contact email: primandari.arum@uii.ac.id

Abstract

In recent years, Bitcoin has attracted a lot of attention because of its nature that supports encryption technology and monetary units. For traders, Bitcoin becomes a promising investment since its fluctuating prices potentially draw high profit (the higher the risk the higher the return). Unlike conventional stock, Bitcoin trades for 24 hours a day without a closing period, so that it escalates the risk. Predicting the value of Bitcoin is expected to minimize the risk by considering some information such as blockchain information, macroeconomic factors, and global currency ratios. However, the multicollinearity among these independent variables causes regression method cannot be used. This research employs Bayesian Regularization Neural Network (BRNN) which is a free assumption. This method is Single Hidden Layer Feed Forward Neural Network (SLNN) that utilize Bayesian concept to optimize weights, biases, and connection strengths. The data is time series data from January 23, 2017, to January 23, 2019. Regression with subset selection is employed to reduce independent variables, from a total of 25 variables to 14 variables. As a result, the predicted value is not much different from the actual data, with an accuracy of 91.1% based on the MAPE value.