0
$\begingroup$

I am running a logit model trying to predict purchases on a dataset including change variables, i.e. I have a dataset of this kind:

              webvisits.month1  webvisits.month2 webvisits.month3 Purchase
contract1          34                   21            22          0
contract2          11                   2             2           1
contract3          9                    22            17          1
contractn          5                    44            42          0

The model is not performing well at all, would it be a good idea to try and normalize my variables? Would that affect the outcome? If this is the case, should I normalize them by month (considering the values by variable webvisits.month1, webvisits.month2 and so on) or rather by contract (e.g. considering the distribution contract1 [34,21,22,0], contract2 [11,2,2,1] and so on? Thanks, hope this makes sense.

$\endgroup$
  • 1
    $\begingroup$ What do you mean with "normalize"? There are many different techniques that could be called that. Regardless, I would expect none of them to really help. $\endgroup$ – Maarten Buis Sep 29 '16 at 13:15
  • $\begingroup$ I think you will find the information you need in the linked thread. Please read it. If it isn't what you want / you still have a question afterwards, come back here & edit your question to state what you learned & what you still need to know. Then we can provide the information you need without just duplicating material elsewhere that already didn't help you. $\endgroup$ – gung - Reinstate Monica Sep 29 '16 at 14:24
  • $\begingroup$ Thank you, differently from the thread you're referring to, I was asking about normalization in logistic regression and whether the normalization is more appropriate across variables or across observations. $\endgroup$ – La Machine Infernale Sep 29 '16 at 18:28
  • $\begingroup$ @LaMachineInfernale, you need to ping me (as at the beginning of this comment), or I won't be notified that you said something. What do you mean by "normalized" here, scaled to lie between 0 & 1? If you can clarify what you're asking, your question may not be a duplicate & may be a good one here. If so, I can re-open this. $\endgroup$ – gung - Reinstate Monica Sep 30 '16 at 0:58
0
$\begingroup$

What you need to improve your model is not normalisation, but create extra features which could affect the target. e.g. captured the change across months in independent variables: webvisits.month2-webvisits.month1 or average, max of 3 months. capture the increasing and decreasing trend. Again just webvisits might not be good predictors, you might need to include other information in model, like what they did in the webvisit. Hope this helps!

$\endgroup$
0
$\begingroup$

Logistic regression does not make the assumption of normality. I doubt the problem lies there.

As a general remark, whatever transformation you do, do it by variable (i.e. the columns from your dataset). Do not independently alter individual rows which are your observations.

$\endgroup$

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