dont mean to to grave dig and while im not trading live yet, i am building a few models to predict direction and how much a stock will move it is very difficult to get an algorithm to predict the next days direction and accuracy isnt important. LIFT is what is important! (how much better those predictions that manage to be correct vs those that are incorrect). ive had people tell me they trade algorithms that score way less than 50% in regards to accuracy but are profitable. a good model will generalise well on unseen data how to craft features and normalize data is important algo/quant funds (the good ones) do not use T.A other than maybe some moving averages and many of them use unique and diseperate datasets that have nothing to do with finance and incorporate them into their models. ML is good at finding correlations that a human wouldn't be able to do