A decision tree algorithm with segmentation

A novel version of the ID3 induction´s decision tree algorithm is presented. The aim of this version is to work with feature values which have a measuring process with noise. The induction input is a table with feature values where objects are labeled. Different classifications may be associated with the same set of features, and the association of conditional probabilities with each classification is discussed. Features are assumed to be either numeric or not. The features model is discussed and it is indicated how the numeric feature values from the initial table are converted into discrete values (nonnumeric). An algorithm where the process of feature segmentation is integrated in the design of decision tree is discussed. The proposed algorithm can be applied directly to sensorial integration in robotics applications