Metrohm Vision Manual - Manual (Page 104)

102 ▪▪▪▪▪▪▪
4.4 Clustering Method Development
Introduction
Library clustering is a new and powerful technique which allows the discrimination between very
similar products in a library containing a variety of different products. It also speeds up the process of
searching a large library for a possible match.
The clustering algorithm initially uses the library global Principal Component model to divide the
whole library into a set of clusters. The user specifies clustering parameters such as the cluster radius
and the distance between clusters. Each cluster contains spectrally similar products and represents a
local library neighborhood in the PC space.
Consequently each cluster may be further divided into sub-clusters. In this case, the PCA is performed
separately for each cluster and the series of PCA models calculated locally is used to define the sub
cluster structure.
This process of subdivision will be continued until the lowest level clusters (so called leaf clusters)
reach a predefined size (minimum of 4 products), or until their radii are too small to be further sub-
divided. The lowest level cluster can have a local identification method, with its own pretreatment
and threshold.
Each lowest level cluster can have a completely local identification method attached to it. Initially one
method common to all leaf clusters is defined together with clustering parameters, and developed
after the library clustering calculations are done. This method can consequently be edited for each
leaf cluster separately.
4.4.1 Library Clustering Procedure
Preparation
1. From the main menu, select Mode/Qualitative Analysis/Library Clustering to enter the cluster
analysis program.