Timeline for What to do when one feature has very large importance/weight?
Current License: CC BY-SA 4.0
7 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 28, 2022 at 12:37 | comment | added | Nicolas Martin | You're welcome Daria. Do not hesitate to validate the answer for data management please :) | |
| Jun 5, 2022 at 8:41 | comment | added | Daria | Yes, will try that. Thank you so much for help again! | |
| Jun 3, 2022 at 18:14 | comment | added | Nicolas Martin | I recommend to check if the logo has actually no correlation with other features, in order to justify that it is not meaningful. Otherwise, you can always double the meaningful features to reinforce their impact, but it is not a very clean method and might differ from an algorithm to another. | |
| Jun 3, 2022 at 16:39 | comment | added | Daria | Yes, negative correlations, and i am not removing them, I was just describing what i have. My problem is the categorical features(like Logo) that have too strong predictive power, Im not sure what to do with them. I guess I will have to just remove them. Thank you again fro help! | |
| Jun 3, 2022 at 13:04 | comment | added | Nicolas Martin | You're welcome Daria. Negative or no correlation? Negative correlations are good, because it is a kind of correlation that may help your predictive model. They shouldn't be removed. | |
| Jun 3, 2022 at 12:14 | comment | added | Daria | Thank you so much for your answer @Nicolas Martin! I actually have more that these three predictors, I just mentioned these three to simplify things. Also these three have the highest correlation with churn. Most of the numerical predictors are activity related, so most of them have negative correlation with churn. I also have a couple other categorical predictors, but they don't influence the predictions and much as Logo. So on one side I want to remove Logo, but on the other side I feel I shouldn't , so I thought maybe there is a way to decrease its coefficient. Or maybe its wrong to do... | |
| Jun 3, 2022 at 7:27 | history | answered | Nicolas Martin | CC BY-SA 4.0 |