Multi-classification method based on support vector machine and containing unknown type

Abstract

The invention provides a multi-classification method based on a support vector machine and containing unknown types; the method comprises the following steps: 1, identifying whether a to-be-identified sample belongs to the i th type or not respectively according to each trained i th type classifier, wherein the i th type classifier is a binary-class classifier obtained by the following steps: using known samples belonging to the i type to build a positive sample set, using known samples belonging to residual N-1 type to build a negative sample set, using a SVM model to train so as to obtain the i type classifier used for identifying whether the inputted sample belongs to the i type or not; 2, inputting the to-be-identified sample in step1 into each binary-class classifier, if all results are negative, the to-be-identified sample is determined to be an unknown type; in step 1, if only one binary-class classifier output result is positive, the to-be-identified sample is then determined to belong to the type matched with the binary-class classifier. The method can detect unknown types, and is high in recall ratio and precision.

Description

technical field [0001] The invention relates to the field of machine learning, in particular, the invention relates to a support vector machine-based multi-classification method containing unknown categories. Background technique [0002] As a typical machine learning algorithm, support vector machine (SVM) has been widely used in various classification problems due to its complete theoretical support. SVM assumes sample data The original space is generally not linearly separable, and the sample data can be mapped from the original space to a new feature space (the new feature space is generally higher-dimensional), and the sample data The corresponding point in the new feature space is The goal of the binary classification SVM can be described as finding a linear hyperplane in the feature space using the sample data in Is the normal vector of the linear hyperplane, b is the offset, if f(x i )>0 then y i =+1 means that the sample data belongs to the positive ca...

Images