Cryptocurrency Price Prediction Using LSTM and RNN Techniques

Abstract

The use of cryptocurrency has been steadily increasing in recent years. Cryptocurrencies such as Bitcoin (BTC), Litecoin (LTC), and Ethereum (ETH) are currently a thriving open-source community and payment network used by millions of people. As the value of these cryptocurrency prices varies every day, technically it would be difficult to predict the buying and selling prices, at the same time investors must forecast the cryptocurrency pricing on a daily basis. Existing techniques for predicting cryptocurrency values use Convolution neural networks (CNN) which are suitable for classification, and might produce suboptimal accuracies in prediction when working with sequential data. Recurrent neural network (RNN) and Long Short-Term Memory (LSTM) are used in this work to obtain more accurate and efficient price prediction. This work considers Mean Squared Error (MSE) matrix as the metric to evaluate the quality of predictions. The RNN algorithm is implemented in the first layer of LSTM. The algorithm is implemented using streamlit library in Python fetching live data from yahoo finance cryptocurrency data. The algorithm produces better accuracy when compared with the existing models.