Tell me more ×
Cross Validated is a question and answer site for statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
up vote 1 down vote favorite
1

I am running a model for a problem in insurance domain. The final results show some false positive x and some false negative y. I am using SAS Enterprise Miner for this. Can somebody suggest me how to reduce false positive? I know for this i have to increase the false negative. I want to know two things:

  1. Is there any option in e-miner where I can give more weight to false negative and less to false positive?

  2. Is there any general approach in modeling which tells us any ways to reduce false negatives or is it just a hit and trial approach?

share|improve this question
Could you please add information what method are you using? – mbq♦ Dec 24 '10 at 9:54
And please add googlable names of the software or at least links to their sites. – mbq♦ Dec 24 '10 at 9:57

2 Answers

up vote 3 down vote accepted

Regarding first (and second) question: A general approach to reduce misclassifications error by iteratively training models and reweighting rows (based on classification error) is Boosting. I think you might find that technique interesting.

Regarding second question: The question sounds a little bit naive to me (but I maybe did not understand your true intention), since reducing misclassification error = improving model performance is one of the challenges in Data Mining / Machine Learning. So if there were a general all-time working strategy, we all would have been replaced by machines (earlier than we will anyways). So I think that yes, the general approach here is educated trial and error. I suggest this question, Better Classification of default in logistic regression, which may give you some ideas for both questioning and model improvement.

I suggest to play around a little bit and then come back to ask more specific questions. General questions regarding model improvement are hard to answer without data and/or additional information of the circumstances. Good luck !

share|improve this answer
up vote 2 down vote

What you can do if you do not find the weight option is create the same effect yourself, by increasing the amount of the positives, for example you can give as an input to the algorithm 2 times each of the known positives an leave the negatives as they where. You can even increase it 10 times, it is a matter of experimenting to get as near as you can to the best possible result.

share|improve this answer
Yes, this can be done if the classification algorithm cannot utilize weights directly (e.g. Naive Bayes is able to). However it is important to note that a mindful validation is absolutely necessary, since this approach may induce a Sampling Bias into the model. – steffen Dec 26 '10 at 20:55
@ steffen -- can u please elaborate on the weight thing..I didnt get it..I dint get the meaning of weights here..does it have any thing to do with weights option in proc logistic..can you please xplain this a bit more ?? thanks in advance. – ayush biyani May 13 '11 at 18:13

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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