Introduction: Knowledge discovery from data (KDD) is a multidisciplinary discipline which appeared in 1996 for “non trivial identifying
of valid, novel, potentially useful, ultimately understandable patterns in data”. Pre-treatment of data and post-processing
is as important as the data exploitation (Data Mining) itself. Different analysis techniques can be properly combined to produce
explicit knowledge from data.
Methods: Hybrid KDD methodologies combining Artificial Intelligence with Statistics and visualization have been used to identify patterns
in complex medical phenomena: experts provide prior knowledge (pK); it biases the search of distinguishable groups of homogeneous
objects; support-interpretation tools (CPG) assisted experts in conceptualization and labelling of discovered patterns, consistently
with pK.
Results: Patterns of dependency in mental disabilities supported decision-making on legislation of the Spanish Dependency Law in Catalonia.
Relationships between type of neurorehabilitation treatment and patterns of response for brain damage are assessed. Patterns
of the perceived QOL along time are used in spinal cord lesion to improve social inclusion.
Conclusion: Reality is more and more complex and classical data analyses are not powerful enough to model it. New methodologies are required
including multidisciplinarity and stressing on production of understandable models. Interaction with the experts is critical
to generate meaningful results which can really support decision-making, particularly convenient transferring the pK to the
system, as well as interpreting results in close interaction with experts. KDD is a valuable paradigm, particularly when facing
very complex domains, not well understood yet, like many medical phenomena.
Presentation slides available from: http://www.bridgingknowledge.net/Presentations/Symp11_Gibert.pdf
Thanks to Luis Salvador-Carulla for trusting KDD. This research has been partially financed by PRODEP (Generalitat de Catalunya), Institut Guttmann and project TIN2004-01368.