Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
7609723 | Journal of Chromatography A | 2017 | 5 Pages |
Abstract
An analysis and comparison of the use of four commonly used error measures (mean absolute error, percentage mean absolute error, root mean square error, and percentage root mean square error) for evaluating the predictive ability of quantitative structure-retention relationships (QSRR) models is reported. These error measures are used for reporting errors in the prediction of retention time of external test analytes, that is, analytes not employed during model development. The error-based validation metrics were compared using a simple descriptive statistic, the sum of squared residuals (SSR) of outliers to the edge of an error window. The comparisons demonstrate that Percentage Root Mean Squared Error of Prediction (RMSEP) provides the best estimate of the predictive ability of a QSRR model, having the lowest SSR value of 20.43.
Related Topics
Physical Sciences and Engineering
Chemistry
Analytical Chemistry
Authors
Maryam Taraji, Paul R. Haddad, Ruth I.J. Amos, Mohammad Talebi, Roman Szucs, John W. Dolan, Christopher A. Pohl,