SIMPLE VALIDATORY TOOLS FOR JUDGING THE PREDICTIVE PERFORMANCE OF PARAFAC AND THREE-WAY PLS

The methods PARAFAC and three-way PLS are compared with respect to their ability to predictreversed-phase retention values.Special attention is paid to simple validatory tools,the meaning and useof which are explained.The simple validatory tools consist of percentages of explained variation in the training set and thosethat can be calculated with the use of markers.These markers are special(reference)solutes,the retentionvalues of which are used to gain information about a new object for which predictions are wanted.Different validatory tools can be calculated with the use of these marker retention values:percentagesof used variation and mean sum of squared residuals after applying the model to these marker retentionvalues.The validatory tools are evaluated on their power to estimate their test set counterparts:thepercentages of explained variation in the test set and mean sum of squared prediction errors in the test set.Two different data sets from reversed-phase chromatography are used to evaluate the validatory tools.The first data set has a high signal-to-noise ratio and is measured under the same measurementconditions.The second data set has a low signal-to-noise ratio and is measured under differentmeasurement conditions.Some of the simple validatory tools seem to have relevance to their test setcounterparts,even in the case of the second data set.

SIMPLE VALIDATORY TOOLS FOR JUDGING THE PREDICTIVE PERFORMANCE OF PARAFAC AND THREE-WAY PLS

The methods PARAFAC and three-way PLS are compared with respect to their ability to predict reversed-phase retention values.Special attention is paid to simple validatory tools,the meaning and use of which are explained. The simple validatory tools consist of percentages of explained variation in the training set and those that can be calculated with the use of markers.These markers are special(reference)solutes,the retention values of which are used to gain information about a new object for which predictions are wanted. Different validatory tools can be calculated with the use of these marker retention values:percentages of used variation and mean sum of squared residuals after applying the model to these marker retention values.The validatory tools are evaluated on their power to estimate their test set counterparts:the percentages of explained variation in the test set and mean sum of squared prediction errors in the test set. Two different data sets from reversed-phase chromatography are used to evaluate the validatory tools. The first data set has a high signal-to-noise ratio and is measured under the same measurement conditions.The second data set has a low signal-to-noise ratio and is measured under different measurement conditions.Some of the simple validatory tools seem to have relevance to their test set counterparts,even in the case of the second data set.