Abstract (EN):
Merging decision trees models has been, so far, motivated as a way to avoid both transporting data sets on distributed locations or training very large data sets. This paper presents a novel rationale which is the need to generalize knowledge by grouping models consumed across different decision levels in a non-distributed environment and propose a methodology for this goal. The approach is evaluated using data from the University of Porto, in the context of predicting the success/failure of students in courses. The experiments focus mainly on the impact of the order of models on the overall performance of merged models. Directions
of unexplored issues for future research are also discussed.
Idioma:
Inglês
Tipo (Avaliação Docente):
Científica