Fisher's decision tree

摘要: Univariate decision trees are classifiers currently used in many data mining applications. This classifier discovers partitions the input space via hyperplanes that orthogonal to axes of attributes, producing a model can be understood by human experts. One disadvantage univariate is they produce complex and inaccurate models when boundaries not axes. In this paper we introduce Fisher's Tree, it takes advantage dimensionality reduction linear discriminant uses decomposition strategy trees, come up with an oblique tree. Our proposal generates artificial attribute split recursive way. The tree induces whose accuracy, size, number leaves training time competitive respect other reported literature. We use more than ten public available sets demonstrate effectiveness our method.