Abstract:

Nowadays, results obtained from classical pre-post studies are based on traditional and often basic,
statistical techniques (see chapter 3). Till now, these techniques have not been expressive enough to
allow the extraction of complex relationships, as the level and degree of interactions between different
subset of attributes is too complex in this context to be captured by simple data analysis or pre-post
statistical testing.
We propose a methodology for analyzing this type of studies which include AI techniques and decompose
the problem in such a way that local multivariate interactions could be detected and modeled.
Clustering methods will be introduced to reduce the original problem space over rows dimensionality
to a simpler set of elements to be modeled, each one concentrating more general conceptual contents.
On the other hand, the attribute space will be also split in different subsets playing different roles in the
analysis according to their different structure and behavior.