I am using 3 features (x1, x2, x3) for regression. Some of my features are continuous some are categorical.

My dependent variable are lets number of bookings. And I can predict the number of bookings. By obtaining how important each feature was in regression as follows (i.e. feature importance)

x1 --> 0.1
x2 --> 0.5
x3 --> 0.7

It is clear that feature 3 (x3) contributes the most, x2 the second and x1 the least in classification.

I also performed correlation analysis to check if my features are positively or negatively correlated with the target (y) as follows.

x1 --> positively correlated
x2 --> positively correlated
x3 --> negatively correlated

I am wondering if it is possible to convert my regression features into a ranking function using feature importance and correlation.

For instance, my suggestion looks as follows.

ranking_score = 0.1*x1 + 0.5*x2 + 0.7*(1/x3)

The reason for using (1/x3) in the above equation is because it is negatively correlated with the target (y). Please let me know if my ranking_score equation is statistically correct? If not, please let me know your suggestions.

EDIT: Why ranking is important to me?

I am happy to provide more details if needed.

  • $\begingroup$ Hi, is this an R related question? if so please consider adding the appropriate tag. $\endgroup$
    – jottbe
    Commented Dec 20, 2019 at 21:16
  • $\begingroup$ No its not an R related question. $\endgroup$ Commented Dec 20, 2019 at 21:22
  • $\begingroup$ I'm a bit confused by your question: you are doing regression on these features in order to predict the target variable "number of bookings", so your model predicts a real number right? If so why don't you rank the values predicted by the model? Also your "ranking_score" is not a rank, it's actually very similar to the linear regression equation (except for the 1/x3: a negatively correlated variable would be weighted with a negative coefficient). $\endgroup$
    – Erwan
    Commented Dec 21, 2019 at 1:38
  • $\begingroup$ I am saying i get the importance of the features based on that i can give weightage to features. And combine them accordingly for a score. What u mean rank by predicted values $\endgroup$ Commented Dec 21, 2019 at 6:50
  • $\begingroup$ @user3432888 It looks like you are trying to redo manually something which already exists, but I'm not sure what: do you want to obtain a ranking, i.e. to order the instances from the first to the last, and if yes according to what? Or do you want to predict a score for each instance, so in this case you should use regression, but you have done regression already, right? Or maybe there is a confusion between the task of classification (predicting a categorical output) and the task of regression (predicting a numerical output)? $\endgroup$
    – Erwan
    Commented Dec 22, 2019 at 22:42

1 Answer 1


There are different ways to interpret your question.

If you are interested in better evaluating the relative order of the features, try permutation importance .

If you are interested in ordering targets, reframe it as a learning to rank problem.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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