LSTM is widely adopted in nlp tasks because it keeps a long memory of sequence data. Given this powerful feature, it can also be used on time series tasks and often found outperform traditional time series methods like ARIMA. On this project, I implemented LSTM on four-years' daily price data of bitcoin and achieved reasonably good prediction results. I also found that multivariate (i.e., multiple time-based features) LSTM models perform better than univariate(i.e., single time-based feature) models.
Data
