Cryptocurrencies Price Prediction using LSTM Neural Network model

Abstract

Cryptocurrencies are rising popularity day by day and attracting more and more attention from both academia and financial industry. Nowadays, digital currencies provide an alternative exchange currency method to the well-known traditional one. Since their use in financial applications is constantly growing, investors are becoming increasingly attracted to this promising type of investment. However, due to the complexity of the temporal dynamics of digital assets, predictions remain difficult to perform. The cryptocurrency market is characterized by high volatility and sharp price swings over a short period of time; hence, the development of effective and reliable price forecasting models turns out to be extremely important for financial investors in order to take accurate decisions. These problems can be overcome by predicting cryptocurrency prices through a machine learning technique. As contribution, this thesis provides an empirical study on applying Long Short-Term Memory (LSTM) model to predict five major cryptocurrencies that are: Bitcoin (BTC), Ethereum (ETH), Cardano (ADA), Solana (SOL) and Polkadot (DOT). The study starts from the data collection, needed for the data analysis process, to the LSTM model evaluation. The accuracy of the model performance is evaluated in terms of Root Mean Square Error (RMSE), Mean Square Error (MSE), Mean Absolute Error (MAE) and R-Squared (R2). These parameters are computed for all five cryptocurrencies to determine in which of these, LSTM model is the best fit to predict accurate prices for the future.