Information Technology Reference
In-Depth Information
Even though SOMs are not primarily constructed to identify and/or
quantify input-output relationships between the causal factors and
response variables, they can be used to analyze this kind of relationship.
It should be noted that the SOM itself does not imply any cause and
effect relationship among the data.
SOMs have been employed as an alternative to the Spearman
product-moment correlation matrix, for correlating the input variables in
studying the powder fl ow (Kachrimanis et al., 2003). The Spearman
product-moment correlation is a nonparametric measure of statistical
dependence between two variables. It assesses how well the relationship
between the two variables can be described using a monotonic function.
Authors have addressed correlation of different micromeritic properties
of pharmaceutical excipients and demonstrated similarity of SOM
component planes for several variables, for example, bulk density and
roundness of the particles, AR, particle size, etc. Even though correlations
observed on component planes cannot be expressed quantitatively, visual
inspection of SOMs is informative and accompanies calculated correlation
matrix of the input variables.
SOMs have been used in characterization of drug release from
SE matrix tablets (Chansanroj et al., 2011), where the training algorithm
followed competitive learning of the neural gas algorithm rule. Neural
gas is an algorithm similar to the previously described Kohonen's,
the main difference being that the map's neurons are not bonded
elastically to each other, therefore being able to freely move in the
feature space. By employing the soft-max adaptation rule of the
Neural gas algorithm, training of the SOM can be simplifi ed (Koga
et al., 2006).
SOMs can be applied for optimization purposes. SOM clustering
was used to divide data into several clusters and identify the cluster
containing global optimum (Arai et al., 2011). It was concluded that
the evaluation method based on bootstrap resampling and a SOM can
be used for the quantitative evaluation of nonlinear response surface
model precision.
5.6 Conclusion
This chapter provided a basic theoretical background of some of the
most commonly applied methods based on artifi cial intelligence,
accompanied with illustrative examples of their application. Benefi ts
Search WWH ::
Custom Search