Incorporating stochastic processes to overcome unmanageable decision trees

Abstract

Many decision problems, when approached by decision analysis in practice, have the property that a full, accurate and realistic decision tree mushrooms into unmanageable proportions. A pruning process undertaken to overcome enormous trees handles excessive width of the tree. This paper outlines an idea for managing the depth of decision trees by partitioning the time horizon into a short and long term. A decision tree of reasonable depth covers the short term, and events that can occur over extended periods of time are represented by a Markov process. Generalization of this idea is possible if another type of a stochastic process is applicable. With the proliferation of end user computing, software support for the idea is now available.