This repository is part of an article about how to forecast and detect anomalies on time-series data. The main objective is to train a RNN regressor on the Bitcoin dataset to predict future values on then detect anomalies in the whole data window - that last step achieved by implementing a RNN Autoencoder. You'll see some other models in the notebooks that I've provided to you in case they are of your interest and this RNN regressor + RNN Autoencoder doesn't perform well for your purpose in any other scenario. The dataset used is available at https://www.kaggle.com/mczielinski/bitcoin-historical-data and contains BITCOIN/USD 1-minute candle data, from 2012-01-01 to 2020-12-31. I hope you can get advantage of this approach!
