Many data sets have samples where a portion of the data is not completely consistent with the behavior of the remainder of the data, but the difference is not sufficient to make these samples outliers in the usual sense. Unusual, but not outlier, behavior may arise from masking, where the unusual systematic effect arises consistently in several samples, but it can also manifest itself as a local, systematic effect that differs in one or a small number of unusual samples relative the the remainder of the data. If the size of the systematic effect in the unusual samples is large enough, the calibration or the predictions from a calibration may be degraded.<br/><br/>We have begun to study methods for finding and correcting for these unusual samples, and will demonstrate the discovery and correction of unusual, systematic effects in spectral data for multivariate calibrations. Methods based on discovery of unusual variables and on discovery of unusual samples will be demonstrated on some real data sets.

Compensating for the effects of unusual samples and variables in data for multivariate calibrations

251st National Meeting (2016)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.