Increase Efficiency and Model Applicability Domain When Testing Options That Are...
Our World Statistics Day conversations have been a great reminder of how much statistics can inform our lives. Do you have an example of how statistics has made a difference in your life? Share your story with the Community!
Increase Efficiency and Model Applicability Domain When Testing Options That Are at First Glance Multilevel Categorical Factors
Mar 1, 2017 7:29 AM
| Last Modified: Mar 28, 2017 1:05 PM
Silvio Miccio, Procter & Gamble
Increase efficiency and model applicability domain when testing options that are at first glance multilevel categorical factors
When testing options of e.g. different raw materials or formulation ingredients, common practice is to vary them as multilevel categorical variable e.g. A, B, C…. in an experiment. Hence, for identifying the best option all of them have to be tested. A consequence of this is
time consuming physical testing is required and the
resulting model is only applicable to predict the tested options but cannot predict options that have not been tested
A much more efficient approach is to design the experiment based on the physical/chemical properties of each option. This,
significantly decreases the number of required experimental conditions and
results in a model that can predicted options not tested before.
The presentation will demonstrate how to:
compress the available information describing the option properties via principal components
select the “corners of the box” for testing representative raw materials based on Design of Experiments.
model the data via PLS
find the overall optimum solution
identify physical available options closest to the calculated optimum solution
Demonstrates the efficiency of this approach
Notice that what is shown here is based on a method commonly used in Chemometrics, called Quantitative Structure – Activity Relationship (QSAR).