Hi dear researchers
I am using R language for implementing a naiveBayes model. My data set includes about 25 factors. The Model and its predicts are fine acceptable. I was wondering how can I find out which one of those factors were more effective in the model.
Thank you very much for your answer
Ali Madad
Since Naive Bayes assumes conditional independence between your factors/covriates, you can test which factor is most discriminative by turning 'off' the other factors, and checking what error you get on the test set.
e.g. with 2 features and 2 classes, NB posterior (ignoring the denomimnator)
P(class1)=P(feature1/class1)*P(feature2/class1)
now, you can calculate 2 terms,
P1(class1)=P(feature1/class1)
P2(class1)=P(feature2/class1)
similarly, for class2:
P1(class2)=P(feature1/class2)
P2(class2)=P(feature2/class2)
now, u can calculate this for your test set, and get the true positive/.false positive rates and check which feature gives better performance.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R