Design of a decision tree with action

Describes an algorithm for the design of a decision tree applied to the identification of objects where the action is built-in. The authors´ main interest is the design of a decision tree with action applied to the sensorial integration problem, for instance a tactile application or vision system with the ability to change the light conditions, but it can be generalized to other applications. The knowledge representation is supported by a graph where each node represents a world state and where the arcs represent the actions of the world. Each world state is composed by a set of knowledge states. The intermediate and terminal knowledge states are in a world state. The intermediate states are defined considering that the knowledge of the possible classification object is incomplete. The terminal state means that the classification is done. Associated to each knowledge state there is the definition of a set of features its values and a training set. The training set is a table with feature values in which classification is known