2
$\begingroup$

I have a stream of data coming in like below (random numbers 0-9)

7, 7, 0, 0, 8, 9, 2, 7, 3, 8, 2, 8, 5, 7, 0, 8, 7, 8, 5, 3, 2, 6, 1, 9, 5, 7, 5, 3, 4, 9, 1, 3, 5, 5, 0, 7, 7, 5, 2, 8, 8, 7, 5, 5, 5, 2, 9, 7, 2, 1, 0, 0, 5, 7, 1, 4, 2, 7, 8, 8, 5, 2, 7, 5, 7, 1, 7, 2, 0, 5, 7, 5, 2, 6, 3, 6, 3, 6, 1, 9, 1, 9, 7, 2, 3, 9, 8, 8, 4, 9, 8, 2, 5, 3, 4, 0, 3, 1, 0, 7, 2, 3, 8, 7, 5, 7, 3, 6, 0, 3, 3, 3, 6, 3, 1, 3, 0, 6, 9, 8, 0, 1, 4, 4, 9, 9, 3, 7, 4, 1, 0, 5, 0, 6, 8, 8, 8, 1, 7, 6

Ask: is to Predict the next numbers(at least 3-10).

Which approach would be helpful in getting through this problem?

$\endgroup$
2
$\begingroup$

Assuming your data is not purely random (otherwise it will be difficult to make useful predictions), you can try the following:

  • Hidden Markov Models: Treating your discrete numbers as states, you can use HMMs to predict the next state (next number) by looking at the last or the $n$ last states.
  • Recurrent Neural Networks: Assuming you have a lot of number sequences, you can treat your numbers as time series. Train a model and predict the next number(s). Here is a nice tutorial which uses LSTM cells.
  • Autoregressive models: AR models are quite powerful as well and have already been successfully used in finance and signal processing before the rise of deep learning.

Further this post (Problem in discrete valued time series forecasting) on Cross Validated might answer your question as well.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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