Machine learning from historic BTC data

by Dan McKeown

Oct 23, 2018

I designed a machine learning algorithm that analyzes a series of hourly Bitcoin prices and then creates a "score" for predicting the next hour somewhere between 0 and 1 (the higher the score, the more likely that the price will be higher in one hour).

A version of this tool is available on the web; however (as that site notes) it is stuck training on older data.

On my computer, I have new data continuously being pulled in from the web (every hour, when the system is running anyway). So it is able to create a new score every hour. Now I have created a Twitter account that posts the hourly score.

This is an example of one of the tweets on this account:

Now predicting the future is an uncertain task ("always in motion is the future") and these predictions may be inaccurate. This is research data right now and I would be hesitant to use it for production purposes.