Dealing with variable certainty in categorical data I have a question regarding statistical analyses for categorical variables. Without complicating it, I would like to know how you can deal with uncertain categorical data. That is, categorical data that have associated confidence levels that range from 1-99%. Here is a screenshot of the spreadsheet the data have been inputed in to.

Are there statistical analyses that can deal with this?
Thanks in advance,
 A: I assume that you are considering the features in the framework of supervised learning. 
You can create for each feature few derived features based on the confidence level.
After that, use feature selection to choose those that serve you most.
The most straight forward way to derive features based on confidence is to set some levels of confidence ( e.g., above 90%, above 80%,...). If the confidence level in a record is high enough, use it. other wise use null.
More advanced ways are binning according to the confidence distribution of each feature (where you can use equal with bin, equal depth bins) or discretisation with respect to the concept (e.g., minimal mutual information loss).
If you'll choose ten thresholds,  then the initial number of features will be ten times larger. Quite high but still OK.
After you will have the set of features, you will need an evaluation function for the feature selection. I guess that the higher the confidence you'll have, the lower the noise you will have. At a given point, you might lose too mush of the signal. At this point, evaluate them against the concept and see how to advance further.
