I'm working on the following missing data problem to learn more about stats, probability, and machine learning, but I'm not really making progress solving it:
I have a group of unordered, non-unique integers. Because of sampling error, there is a small probability (say <= 5%) that one or more integers is missing. I used a pattern matching algorithm against a training set and found that within each group there is one or more patterns of integers that may overlap. For example:
group with possible missing data: 1, 2, 3, 3, 4, 5
pattern: 1, 2, 3
pattern: 1, 2, 3, 3
pattern: 3, 4, 5
pattern: 3, 4, 5, 6
I have tried a number of classification, clustering, and raw probability approaches using the patterns, but haven't had much luck. My question is, what is the best approach to determine if there are missing integers or what they may be? Are there different strategies?
I'm starting to think I picked too hard of a problem to learn on. :)
