Abstract
A new KDD methodology specially designed to analyze repeated very short serial measures called Knowledge Discovery in Serial Measures (KDSM) is introduced.
In domains where serial measures are repeated and very short (i.e., very few parameters), traditional methods for series analysis are inappropriate. Moreover, some information is non‐serial but is closely connected to serial measures. For this reason, it is necessary to find a suitable way of analyzing these situations. Such an objective is reached with the use of KDSM.
The use of KDSM yielded some very interesting results in each of the domains where it was applied (e.g., psychiatry and labor domain). The analysis of these case studies' data with Artificial Intelligence techniques or Statistics in isolation would never have provided such relevant results.
While analyzing problems inherent in domains related to the context of serial measures, we found that according to most experts in these domains, it is practically impossible to obtain significant results by using isolated methods or techniques due to the singular structure of these domains. Moreover, if it were possible to obtain some results, they might well be too difficult to be interpreted by the experts. Thus, the main goal of this thesis work is to overcome the limitations of Artificial Intelligence or Statistics techniques.
A hybrid methodology which uses the best combination of other more simple techniques, thus facilitating the study of this kind of domains, was chosen.