Abstarct: Cryptocurrencies, considered digital or virtual currencies secured by encryption technology, have attracted widespread attention in economics and investment. Their decentralized nature, free from government control, has attracted both individual and institutional investors since Bitcoin emerged in 2009. The unpredictability of cryptocurrency prices creates challenges for investors, institutions, policy makers and regulatory bodies. Time series analysis and heuristic algorithms, including artificial neural networks, are commonly used for price forecasting. In this research, the closing prices of three cryptocurrencies, Bitcoin, Ethereum, and Litecoin, were first modeled and predicted by ARIMA and SARIMAX statistical models. The MAPE error of these two methods was obtained for each cryptocurrency, which outperformed the overall ARIMA model. Next, the limitations of statistical models were examined, and this was the beginning of addressing more advanced methods. In order to predict the price of cryptocurrency by different artificial neural network architectures, a data window was first created, which categorized the data in a way like time series to enter the model. Then, the performance of linear, deep and LSTM models for one step and several forward time steps and ARLSTM for several forward steps on three cryptocurrencies including Bitcoin, Ethereum, and Litecoin in the crypto market were investigated and the predicted outputs were displayed in the form of a data window. Evidence and results showed that LSTM neural network for one time step and ARLSTM for multiple time steps performed better than other models.