A Probability-Based Approach to Soft Discretization for Bayesian Networks

摘要: This report discusses how soft discretization can be implemented to train a discrete Bayesian Network directly from continuous data. The method consists of step that converts the variables training cases into evidence, followe d by suitable parameter learning algorithm for Network. is modificati on Maximum Likelihood Estimation which modified accept evidence as input. We also discuss use inference and convert results network meaningful output values. Most literature Bayesi an Networks proposes fuzzy set theory based membership functions. Our approach goes back one further starts out with probability density function spreads influence variable its neighbors, followed step. Thus our discreti zation theory, rather than theory. then show interesting connection between these approaches. Namely, generated through convolution, yielding probability-based Prime applications this include any system limited data whose underlying mechanism in nature. These types are common natural sciences medicine. Using continuity system, i.e. fact t hat neighboring states related each other, we hope c yield more robust accurate models small sample sizes. describes enough detail allow anyone implement it themselves. Preliminary tests indicate increased robus tness, but extensive performance new comparison traditional have yet performed.