Context

This dataset was inspired by Mr. Zielak's Bitcoin Historical Data. I decided that it would also be good to add more cryptocurrencies, and I did it as best I could. Also in this dataset, there is a dataset for binary classification tasks.

Content

The creation of a dataset is described in detail in the article on the arXive, LINK. It contains data for 3 years, from June 1, 2018, to June 1, 2021. 4 currencies are taken for 1, 5 and-15 minute intervals (what I could find in the public domain). There is no per-minute data in the dataset for bitcoin because Mr. Zielak more than fully presented it in his work. The data is given in the following format: unix,date,symbol,open,high,low,close,Volume USD,Volume EOS (Instead of EOS, the name of the respective currency). Timestamps are in Unix time. There are time periods during which there was no trading activity. I decided not to mark them as NaN, and left them in the dataset. Among the files, there is also a file with data 0 and 1, named "Binary dataset". It is created from a combined file with one-hour data, where 1 corresponds to a situation when the current close price is higher than the previous one, and 0, when it is lower. I hope this dataset for binary classification can be useful for use in models such as LSTM.

Acknowledgments and Inspiration

First, I want to thank Mr. Zielak for the answers to my questions and advice, as well as the inspiration. Secondly, I hope that these data can facilitate research and experimentation with various neural networks. Thirdly, if you will use this dataset for your scientific activities, I would be grateful to you for citing my article, which describes the creation of this dataset (Link will be added soon).