LEARNING METHOD FOR BINARY CLASSIFIER CLASSIFYING SAMPLE INTO FIRST CLASS AND SECOND CLASS

摘要

PPROBLEM TO BE SOLVED: To provide a binary classification method based on a hyperplane discriminant for simplifying an optimization task. PSOLUTION: A set of training samples is acquired, and the respective training samples are labeled to be classified to either of first or second class. A pair of binary samples are connected together by a projective vector so that a first sample of each binary pair belongs to a first class and a second sample of each binary pair belongs to the second class. A set of the hyperplanes is formed so that the set of the hyperoplanes is provided with a plane orthogonal to the projective vector, and one hyperplane minimizing a weighted classification error is selected from the set of the hyperplanes. The set of the training samples is weighted according to the classification based on the selected hyperplane. The selected hyperplanes are connected together and turned into a binary classifier, and when selection and weighting are repeated predetermined times, a final classifier classifying a test sample into the first and second classes is obtained. PCOPYRIGHT: (C)2004,JPO