Time-Series Real-Time Bitcoin price prediction using LSTM

Abstract

Bitcoin, the biggest decentralized cryptocurrency till date, is an extremely volatile digital currency in which new forms/units of currency are generated by the mechanical/computational solution of mathematical problems. In recent years its popularity can't be ignored as it is very significant and evolutionary in its own way and its price has shown high volatility. Due to its volatility and instability of prices, there is a high risk of investing in it if ones don't have any knowledge about it. Bigger players can manipulate the market or they barely go into losses. But for smaller players, it's a high chance to lose their hard-earned money. Therefore, it has become necessary to develop an efficient and more accurate predictive model for forecasting bitcoin prices. Accurate prediction of its price is of great interest to investors and traders as well. In this research paper, Long Short-Term Memory (LSTM), a type of Recurrent Neural Network (RNN), has been used for predicting Bitcoin time series data. We have used real-time data on an hourly basis. Bitcoin, a decentralized digital currency, We collected and pre-processed historical Bitcoin price data from the coin desk, an API for cryptocurrency live data dashboard, and split it into training and testing sets. We have designed an LSTM model and trained it on the training data. We evaluate the performance of the model on the testing data and use it to make predictions on new Bitcoin price data. The experimental results of our experiment demonstrate that the LSTM model outperforms other traditional time series models and provides a promising approach for Bitcoin price prediction with 96.88% accuracy for new data.