In the world of traditional securities, such as a company's stocks, it used to be humans who would do the analytics, predict the prices of stocks, and trade. Today, the development of machine learning  (ML) and the growing availability of data has almost eliminated humans from high-frequency trading, as a regular person can't capture and process all data, and emotions affect one's decisions; so it's dominated by automated trading systems by investment institutions.

Currently, the volume of Bitcoin trading is relatively low compared to traditional exchanges; financial institutions, being traditionally careful and risk averse, haven't got their hands on Bitcoin trading yet (at least, it's not well-known). One of the reasons is high fees and uncertainty regarding regulations of cryptocurrencies.

So today, mostly individuals buy and sell Bitcoins, with all the consequences of irrational behavior connected to that, but some attempts to automate Bitcoin trading have been made. The most famous one was stated in a paper by MIT, and another one was by Stanford researchers, published in 2014. Many things have changed, and taking into account the massive Bitcoin price increase during these three years, anyone who just buys and holds on would be satisfied enough with the results:

Definitely, some traders use ML for trading, and such applications look promising. So far, the best possible approach that was identified from research papers is as follows.