In addition to the efficiency evaluation of data clustering processes supported by MAM built with these new policies, we also evaluated their effectiveness (see 'Analysis of data clustering' section).
In this set of experiments, we employed the PAM-SLIM algorithm with the aim at analyzing the impact on the behavior of data clustering processes when employing different node split policies on the construction of the MAM that supports the clustering algorithm.
However, utilizing only the microarray data in the clustering process resulted in partitioning of the samples into just two groups with R' = 0.51.
Equal weights of the domain data in the clustering process resulted in k = 3 and R' of 0.64 when the four centrilobular necrosis of the liver histopathology observation levels were used as an external indicator of clustering validity.
This process generated a total of 23 basic themes, using Attride-Stirling's [ 36] method of interpretative thematic analysis, a data sorting and clustering process which identifies descriptive basic themes (text segments), which are then grouped into increasingly abstract higher order organising themes and highest order global themes.
Our method is one of a class of approaches that seek to incorporate biological data directly into the clustering process [ 9, 14].
In this paper, we regard the feature selection as a clustering process with data decomposition technique and propose a novel feature selection method based on the non-negation matrix factorization (NMF).
These data are compatible with a clustering process of GM1 molecules, an effect that not only increases the Aβ peptide affinity, but also causes a pronounced Aβ peptide penetration deeper into the lipid membrane; all these factors are potentially involved in Aβ peptide aggregate formation due to an altered ganglioside metabolism found in AD patients.
For B. plicatilis publicly available data from dbEST and the trace archives were included into the clustering process, and public data for Philodina roseola was clustered the same way.
Then, the numerical approaches to determining membership functions proposed in part I of this paper are used to study the fuzzy description and data clustering problems by mimicking human reasoning process.
