PREDICTIVE MODELING IN CRYPTOCURRENCY: A CASE STUDY OF BITCOIN USING RNN

Abstract

Predictive Modeling in Cryptocurrency: A case Study of Bitcoin using RNN is a research based project, which contains the prediction result of the bitcoin prices using the deep learning model trained using LSTM (Long Short Term Memory) model under the RNN(Recurrent Neural Network) system. The data-set used to train the RNN model was composed of a time-series based price of the bitcoin in minutes time frame. Collectively, the size of dataset are over 5,000,000 input points for the two year time-period. The project also consists of a interactive implementation of LSTM model building using different optimizer function, loss Function to compare the result using different models. Addi tionally, the varying factor of Neurons number and Epochs iteration can also be altered using the interactive interface. The problem among other systems was to predict the prices of the hyper volatile time varying component such as Bitcoin Prices using large data-sets. Comparison of accuracy difference using different optimizer function using r2 score. We have simplified the LSTM implementation using various layer of abstraction to make it easy to change various components such as activation and loss function in the model building, training and final prediction result. The outcome is an interactive, ready to use LSTM based RNN model where the type and size of the data sets along with activation and loss function can be easily altered to see varying prediction result.