Boot strapped model reduction steps
After obtaining the full model (all features included):
1. Sampling with replacement (using sampling size the same as the number of total samples).
2. Predict the probability using, for example, 1 feature and its associated averaged weight (in this case this feature would be the one with highest absolute average weight).
3. Evaluate P values against the ground truth of the sampled observations.
4. Repeat these steps 100 times and get the median p value performance (each time the sampled observations in step 1 would be different).
5. Repeat these steps for 2, 3, 4, 5 features and so on.
Do you have any questions about this protocol?
Post your question to gather feedback from the community. We will also invite the authors of this
article to respond.