I stumbled upon this very cool blog article on Investor.blog about how to build a stock trading system with Machine Learning.
The author has spent much time recently on how to build and test a stock trading strategy using Machine Learning. At the end of the research, the author found it quite an impossible task, however it is a lot of fun in the process and sometimes it can even be profitable.
The article offers detailed explanation and source code of the trading system. It first took a look at the stock market, and formulates several trading strategies. Then it built the first strategy using a support vector regression and the second using dynamic time warping.
Hopefully, this will serve as a jumping point for your own explorations.