The Altair Community is migrating to a new platform to provide a better experience for you. In preparation for the migration, the Altair Community is on read-only mode from October 28 - November 6, 2024. Technical support via cases will continue to work as is. For any urgent requests from Students/Faculty members, please submit the form linked here
"Maximum theoretical accuracy"
Maven
Hello,
I've been wondering, is there a theory on the maximum accuracy one can achieve on a given example set without over-fitting? I'm looking for something that could tell me, "whatever method you use to do regression/classification on this dataset, you'll never achieve over x% accuracy" , or something along those lines.
I tried a quick Google search, but didn't find anything along those lines, so I'm asking the experts now
Thanks for any leads on the subject!
I've been wondering, is there a theory on the maximum accuracy one can achieve on a given example set without over-fitting? I'm looking for something that could tell me, "whatever method you use to do regression/classification on this dataset, you'll never achieve over x% accuracy" , or something along those lines.
I tried a quick Google search, but didn't find anything along those lines, so I'm asking the experts now
Thanks for any leads on the subject!
Tagged:
0
Answers
there is a very popular upper bound for the maximum accuracy you can reach without looking at the data: 100%
Sorry, but without knowing anything about the dependencies between the label and the attributes, you simply cannot make any reasonable assumption about the maximal achievable accuracy. Since you are always looking only at a sample of the underlying statistical distribution, you even cannot say with 100% certainty, if the accuracy of your learned model, estimated by a cross validation, is as high as it will be on new, unseen data.
Greetings,
 Sebastian