0
$\begingroup$

So I made a machine learning model which predicts stock movements which returns 1 for the price going up or 0 for the price going down. without the post processing the training accuracy and testing accuracy is in the range of 50% - 54% but if I add the postprocessing the testing and training accuracy goes above 75% depending on how long I train the model and the data i use to train and test it like RSI and MACD.

Here are the logs in the form of the three different post processing techniques:
On Training data:

[0.5000609533097647, 0.8257954406924296, 0.532244300865537]
8203
8202
[0.4964037547238815, 0.8203096428136047, 0.5323662074850665]
8203
8202
[0.49676947458246984, 0.8121418993051323, 0.5323662074850665]
8203
8202
[0.49737900768011706, 0.7832500304766549, 0.5323662074850665]  

On testing data:

2021
[0.5121227115289461, 0.8090054428500743, 0.5314200890648194]
2020
[0.5096486887679367, 0.8169223156853043, 0.5319148936170213]
2021
2020
[0.5091538842157348, 0.8213755566551212, 0.5329045027214251]
2021
2020
[0.5091538842157348, 0.830282038594755, 0.5338941118258288]

As you can see the second technique works best. I just want to make sure that it is ok to do this and this and note that it worked repeatedly on multiple runs with different window sizes and data.

EDIT:
Just to be clear this works on different data and windows unless the validation accuracy stagnates at the start of training which results in an accuracy less than 50.

$\endgroup$
1
  • 1
    $\begingroup$ This completely depends on the what type of post-processing you are applying and how you are applying it, and since this it not mentioned in your question we cannot answer if what you're doing is fine or not. $\endgroup$
    – Oxbowerce
    Jan 1 at 13:41

Your Answer

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

Browse other questions tagged or ask your own question.