Discretization of a Continuous Frequency Value in a Model of Socially Significant Behavior

Abstract

Frequency or behavior rate is one of the main characteristics of behavior and can be defined as the average number of episodes of behavior that occur over a period. Knowledge of behavior rate can be used in many applications to predict behavior and estimate other related properties. Previously, a model based on a Bayesian belief network was presented to estimate behavior rate using data on recent episodes of behavior and the minimum and maximum intervals between episodes. Because Bayesian belief networks involve dealing with discrete values, the model uses discretization of continuous values. In this paper, we examine how different methods of discretization of a continuous variable describing the behavior rate affect the effectiveness of this model’s predictions.