So, i just start learning in AI field and i want to develop my skill more. So, i just curious is it possible to predict last 3 digits of a closing stock price? which mean the output result will run from "000" to "999" in total 1000 number and to make it more clear lets say the closing price of Dow Jones Industrial Average on 06/02/22 is "33248.28" So, it should predict the "828" to be correct in this case. I understand that predicting last 3 digit of a close stock price is guessing and kindna impossible. So, to make the project more practical lets say the code can predict out 500 out of 1000 number. if in the 500 number that it predict out is the last 3 digit of the close price it win. So, i want to know that in your opinion is it practical or possiable or not ? if you think it pratical can i have a suggestion on where should i start or look to? for example like which model will work best in your opinion.

thankyou for answering

  • 1
    the question is not whether you can predict it, but what it is related to. for example you could do some machine learning to see if the the price was related to the day of the week, or the weather or the price of beans 1 week earlier etc
    – Ewan
    Jun 3 at 9:40
  • 1
    @Ewan That's just a different side of the same coin: if the price was correlated to the day of the week, you could predict an output based on the day of the week as input. If you're just identifying the patterns but not applying them, that's not really machine learning, it's just data mining.
    – IMSoP
    Jun 3 at 10:03
  • 1
    the point is you need to throw in more data than just your target predictor
    – Ewan
    Jun 3 at 10:08
  • @Ewan Ah, yes, indeed. Of course no matter how much extra data you throw into the model, you're never going to predict next week's lottery numbers.
    – IMSoP
    Jun 3 at 10:23
  • 3
    ..or are you?!?! for example people have had good results correlating roulette numbers with roulette wheel. The point is you have to have a theory to test, it might be totally possible to predict the stock price based on phase of the moon or something, but if you never add that data to your set you will never know
    – Ewan
    Jun 3 at 11:49

1 Answer 1


AI isn't magic.

Machine learning algorithms work by detecting patterns in a particular set of data, then using those patterns to classify new samples, predict future members of the pattern, etc.

If there's no pattern there to spot, there's nothing for a model to work with. If you try to train a machine learning model to predict next week's lottery numbers, it's never going to work (or, if it does, it will be a major scandal, because the lottery is being rigged or at least very badly run).

The least significant digits of a varying signal are actually a common source of randomness precisely because there's generally no pattern to them. The Dow Jones may not be as explicitly random as lottery numbers, but I'd be very surprised to find any pattern in the last few digits which a machine learning model could take advantage of.

  • 3
    In other words, trying to "predict" even 2/1000 of these numbers would be trying to "predict" 2/6 of every dice roll. It would be just guessing, which means in the long run it would be impossible to get more than 1/1000 and 1/6 correct respectively.
    – Jasmijn
    Jun 3 at 9:46
  • 2
    @Jasmijn Precisely. Another analogy would be roulette: making 500 guesses rather than one is like betting on red or black, rather than a single number, but if the table is fair you're still just guessing.
    – IMSoP
    Jun 3 at 9:52

Not the answer you're looking for? Browse other questions tagged or ask your own question.