Biomedical Engineering Reference
In-Depth Information
K
M
K
K
M
=
+
X
e
S
S
S
Fig. 8.1.
A key idea in chemometrics is to record
K
variables (including spectral
data) for
S
samples to form a matrix
S
×
K
. The critical characteristics of both the
samples and the spectral data can then be understood using a smaller set of matrices
S
×
M
and
M
×
K
while the unmodeled residual remains available for analysis as
the matrix
e
. While the reduced matrices provide insight into the system or process,
residual data can be used to understand errors or limitations of the model
recorded data are repeated measures of a small set of sample characteristics.
The small set defines almost entirely the sample as well as its characteristic
spectral properties.
A number of excellent references are available for classical techniques, in-
cluding those by Kowalski et al. [7, 8], Kramer [9], Brereton [10], compilations
[11, 12], and series of periodic reviews [13]. There are two somewhat related
problems in chemometrics that are relevant to spectral data. The first is to
predict molecular identifications in mixtures from a spectrum and the sec-
ond is to predict a class or label for a spectrum. While regression techniques
work well for molecular identifications, problems of biomedical interest of-
ten involve stochastically varying compositions and spatial distributions of
Data Acquisition
Pre-processing
Feature Selection
Classification
Supervised
Unsupervised
Bayes
Classifier
Nearest
Neighbor
Decision
Trees
Genetic
Algorithms
ANN
LDA
K-means
Hierarchical
Fig. 8.2.
Overview of data processing steps and classification methods discussed in
this chapter
Search WWH ::
Custom Search