1
$\begingroup$

On this site of Otto Group Product Classification Challenge, it is shown that best accuracy was possible with RandomForest method, but it was relatively low at 0.83. Accuracy with ANN and with Naive Bayes was even lower (0.72 and 0.65, respectively).

What is the cause of low accuracy achievable in some data? Are there any other methods with which accuracy can be increased here?

Edit: On this site: https://www.kaggle.com/c/otto-group-product-classification-challenge/leaderboard best model seems to have a score of 0.38. What will be corresponding accuracy for that?

$\endgroup$
2
$\begingroup$

I can mention two main reasons.

  1. Complexity of dataset
  2. high Bayes error

The former means your dataset is highly non-linear and the latter simly means you may have same input with contradictory outputs in your dataset.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Has higher accuracy been reported for this data anywhere? $\endgroup$ – rnso Sep 30 '18 at 14:29
  • 1
    $\begingroup$ I have not tracked the problem but there is a point. If the low accuracy is due to high Bayes error, the accuracy cannot be better. $\endgroup$ – Media Sep 30 '18 at 15:10
  • $\begingroup$ Pl see edit in my question above. $\endgroup$ – rnso Sep 30 '18 at 15:17
  • 1
    $\begingroup$ The reported error value depends on the cost function used. You have to choose an evaluation criterion and find the accuracy. I have to refer to something. In some occasions, even $60$% accuracy is highly acceptable, for instance, handwritten OCR, while for images which are taken from computer made fonts, even $99$% may not be appropriate. $\endgroup$ – Media Sep 30 '18 at 16:00

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.