So, one thing that occurs to me, is that it took a long, long, long time before weather prediction models could improve on "it will be about the same as yesterday". So, you might want to test your code on something like, say, "it will move the same amount as yesterday, in the same direction" and see if that also gets 90%. Not saying it would, just that this is a thing you should check.

I trained my own machine learning model to predict the direction a stock would move the next day using a few things like past prices, volumes, EPS, historical profits, etc.

To check it’s accuracy I generated a prediction for everyday for the past year. Then I checked to see if the price went up from the previous day and I also check if that’s what the model predicted. For some crazy reason, the AI is predicting the movement of some stocks at 90%+ for high confidences (GOOG, AMZN, etc.).

I feel like my math is wrong somewhere but I am not sure. The app that has this is called Hubble Investing for iOS. If anyone checks the app out, let me know the email used, and I can give a free premium account to inspect the AI outcome.

I can also link to a GitHub repo, if anyone wants to check my math. I just feel like it’s not possible to have an AI that predicts so accurately!

Why does the submission link not point to your git?

Fix the title lest your post get flagged and die a sad death..

Because the repo is private rn. I can create a public one with just that section of the code though, if you wanted to check it out!

Sure, please do.

Okay give me like 30 minutes!

Oh, I didn't realize it was your app (did you edit the post? Otherwise sorry for my bad job at reading).

It sounds cool, but what kind of underlying "AI" math are you using?

No worries! To train the model I utilized Turicreate and tensorflow.