Rule flow learning: A multiple linear classifier algorithm
Abstract
Rule flow is a directed graph with condition and action operator over business object's attributes. The results from the the rule flow is usually not linearly separable, which proposes great challenges to rule flow learning from sample results. This paper proposes to use multiple linear classifiers for rule flows whose condition is the linear combination of business object attributes. This is a two-step process. First, to construct the boundary of each category based on the nearest distance points policy. Then, use a stochastic selection approach to approximate the boundary by linear equations. The computation complexity of the process is quadratic level. The feasibility of such process is illustrated by a simple toy sample and air cargo load planning case. ©2009 IEEE.