An SVM classification method based on a hybrid kernel function

Abstract

The invention discloses an SVM classification method based on a mixed kernel function, and the method comprises the steps of 1 collecting a data set, analyzing each sample in a collected data set record, distinguishing different attributes of the samples, and determining input and output samples; 2 selecting and constructing a kernel function, and mixing the exponential distribution kernel function with the radial basis kernel function; 3 optimizing parameters in the mixed kernel function; 4 selecting C-SVC model, establishing a support vector machine classification model based on a novel mixed kernel function; and 5 performing classification prediction through the established support vector machine classification model. According to the method, the global performance of an exponential function and the local performance of a radial basis function are fully utilized, the model parameters are optimized by adopting a particle swarm optimization algorithm with Gaussian variation, and the overall performance of the support vector machine is improved. The learning and generalization performance of the global index distribution kernel function is higher than that of other single kernel functions, and the performance of the support vector machine of the novel hybrid kernel function is obviously better than that of the support vector machines of other hybrid kernel functions.

Description

technical field [0001] The invention relates to an SVM classification method, in particular to an SVM classification method based on a mixed kernel function. Background technique [0002] SVM (Support Vector Machine) refers to a support vector machine, which is a common discriminant method. In the field of machine learning, it is a supervised learning model, usually used for pattern recognition, classification and regression analysis. The SVM method maps the sample space to a high-dimensional or even infinite-dimensional feature space (Hilbert space) through a nonlinear mapping p, so that the non-linearly separable problem in the original sample space is transformed into a feature space in the feature space. linearly separable problems. Simply put, it is dimensionality enhancement and linearization. Ascending the dimension is to map the sample to a high-dimensional space. In general, this will increase the complexity of the calculation and even cause the "curse of dimensi...

AI Technical Summary

Problems solved by technology

If the spatial dimension of the data projection is large, the obtained model is more complex, the empirical risk is small but the confidence range is large, and over-learning is prone to occur; vice versa

[0005] At present, there is no unified rule for the selection or construction of kernel functions. Generally, empirical methods are used to select kernel functions. As long as functions that meet the Mercer conditions can be selected as kernel functions in theory

Images