Potential function methods: Efficient probabilistic approaches to model complex data distributions

Abstract

Potential function methods are a family of non-parametric probabilistic techniques that estimate a probability density distribution of a class of interest as a sum of contributions from each single sample of the class. The resulting estimated overall probability distribution can be sectioned at different confidence levels, with isoprobability contours representing the boundaries of the class space at each confidence level. Boundaries of the class spaces can be very complex, capable of effectively describing non-normal and non-uniform sample distributions.