Techniques for detecting temporal process variation and for managing and predicting performance of automatic classifiers applied to such processes using performance estimates based on temporal ordering of the samples are presented.