Hybrid classification - using axis-parallel and oblique subdivisions of the attribute space

This paper describes the hybrid algorithm DIPOL-DT which uses the strengths of standard decision tree algorithms and piecewise linear classifiers because at every level of learning it chooses the appropriate subdivision of the attribute space: a split with hyperplanes in general position or an axis-parallel split. The proposed method combines two existing algorithms - the piecewise linear classification algorithm DIPOL and the decision tree algorithm CAL5. To some extent the strength of each individual method complements the weaknesses of the other.