Power calculation for 5-level classification problem

Question

I'm trying to do a power analysis in order to determine the sample size required. I have samples belonging to 5 different groups and would like to estimate how many samples in each group I would need in order to develop a classification algorithm based on 5 features for each sample.

So sample A can belong to groups 1 to 5, and we would have measurements for 5 variables for each sample.

Is there a way to easily implement this in JMP?

Thanks

Agustin · Answer

Thank you, I think the problem needs to be defined better first. I'll have a read through the empirical power.

Mark_Bailey · Answer

Power is the probability of correctly rejecting the null hypothesis. I am not sure how that definition applies to classification.

You are likely using the JMP scripting language to implement your classifier. You could also script a Monte Carlo simulation to re-sample while varying the sample size and the feature distributions of each group. Save the classification performance of each iteration and set of conditions. This method is called the empirical power.