Browse Prior Art Database

Optimal Splitting Algorithm for Arbitrary Random Predictors in the Two-Class Problem

IP.com Disclosure Number: IPCOM000036900D
Original Publication Date: 1989-Nov-01
Included in the Prior Art Database: 2005-Jan-29

Publishing Venue

IBM

Related People

Authors:
Nadas, A Nahamoo, D [+details]

Abstract

Disclosed is an algorithm for finding the optimal split in the sample space of a continuous random variable. If (X,Y) has a distribution such that Y has two possible values and X is continuous (e.g., a multivariate Gaussian mixture), then for predicting Y the best question about X cannot be found either by exhaustive search or by sorting the conditional probabilities P(Y = 1 X = x) as in the case of X with finitely many values. An algorithm is given to define the optimal split (question) in the continuous case.