Leveraging Artificial Intelligence to Build Algorithmic Trading Strategies 

Hi, Can you share the URL which you are referring to?

@@quantinsti :: Trying to register over here -- inovancetech.com/#/register .........However, as JustSirk says, the website does not allow login or registration

Yes, machine learning is also applied on order flow prediction and you can google for research papers on the same.

you could put some links that are interesting to know what I have to look for, thanks

Here are some links to research papers on the same Modeling High-Frequency Order Book Dynamics with Support Vector Machines diginole.lib.fsu.edu/islandora/object/fsu:185196/datastream/PDF/view Machine Learning for Financial Market Prediction discovery.ucl.ac.uk/1338146/1/1338146.pdf If you are interested in learning Algorithmic trading we offer a very extensive course called Executive Programme in Algorithmic Trading (EPAT™). You can contact us (www.quantinsti.com/contact-us/) for the course details.

@@quantinsti Hi, the Innovance site for backtesting the signals in this video is non-functional. The signup link does not work. Do they have another site or are there similar sites for backtesting and generating their code??

I dont think order flow will be a good input to predict stock price. Order Flow could be used in sentiment analysis.

+Antonio Bodig We are able to limit the selection bias (the issue described in both sources) by leveraging the human component when selecting the indicators used in the strategy. Instead of generating thousands, or millions, of random variables to be used as the input, as is done in both papers, the user selects indicators that make sense to him/her. This limits the search space to only a few strategies and we can have a lot more confidence in our out-of-sample results as the strategy has a theoretical basis. Our focus is making the process as easy as possible to see if the selected indicators can form the basis of a profitable strategy. While the platform can be misused to create thousands of random strategies, the combination of limiting the search space and complexity of the strategies, as well selecting the indicators based on a conceptual understanding, allows us to minimize the risk described in the above sources.

Out of sample testing & confusion matrix is made to test for data mining bias right ?