Eric Bax

Validation of

This paper presents a method to compute probably approximately correct error bounds for k-nearest neighbor classifiers. The method withholds some training data as a validation set to bound the error rate of the holdout classifier that is based on the remaining training data. Then, the method uses the validation set to bound the difference in error rates between the holdout classifier and the classifier based on all training data. The result is a bound on the out-of-sample error rate for the classifier based on all training data.

k

We introduce a technique to compute probably approximately correct (PAC) bounds on precision and recall for matching algorithms. The bounds require some verified matches, but those matches may be used to develop the algorithms. The bounds can be applied to network reconciliation or entity resolution algorithms, which identify nodes in different networks or values in a data set that correspond to the same entity. For network reconciliation, the bounds do not require knowledge of the network generation process.