1
$\begingroup$

Possible Duplicate:
Supervised learning with “rare” events, when rarity is due to the large number of counter-factual events

I am trying to predict diabetes using the BRFSS dataset by using a supervised learning classification model. But I see that the target variable which is having diabetes or not is skewed. That is 90% of the records are non-diabetic and only 10% of the records are diabetic. How do I handle the skewness in the target variable?

$\endgroup$
  • 2
    $\begingroup$ Why do you perceive "skewness" as a problem that needs correction? $\endgroup$ – whuber Apr 20 '11 at 18:50
  • 2
    $\begingroup$ This question sounds rather similar (stats.stackexchange.com/questions/9398/…) and Dikran gave a good answer to it. $\endgroup$ – steffen Apr 21 '11 at 5:57
1
$\begingroup$

When your data is skewed you may:

  • use specific error metrics like precision, recall, F-score
  • trade of between precision and recall accordingly:
    • want to predict diabetes with confidence => adjust for higher precision, lower recall
    • want to avoid missing too many diabetes cases => adjust for lower precision, higher recall
    • (for example, in logistic regression, by adjusting the separating threshold)
  • use F-score to find a good balance between precision and recall, that maximizes both as much as possible
$\endgroup$

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