Research And Application Of FCM Based Multi-Kernel Support Vector Machine

As one of the research methods of machine learning,support vector machine has been studied in depth and widely used,but support vector machines based on a single kernel function are no longer suitable for applications under complex large-scale data,so multiple kernel support vector machines have become a research hotspot.This paper introduces the principle of multiple kernel support vector machine and the research results of improving the learning efficiency of multiple kernel support vector machine.Based on the clustering of weighted multiple kernel SVMs,the appropriate kernel functions are selected according to the source of the features,and then each kernel function is fitted linearly for different feature sources.A multiple kernel SVM algorithm based on fuzzy clustering is proposed.The experimental results show that the simplified multiple kernel support vector machine based on fuzzy clustering can significantly improve the learning efficiency of multiple kernel SVM,especially on large-scale training data sets.Fuzzy C-Means clustering algorithm(FCM)is one of the popular clustering algorithms of data analysis.In order to solve the problem of traditional FCM clustering algorithm for non-European data structure,there are already many solutions.This paper first selects the kernel function based on the feature source,and the corresponding feature set forms the kernel function set.For better clustering,an adaptive algorithm is used to determine the weight of the kernel function set.In the objective function,the kernel function set is introduced to form a new kernel function-based fuzzy C-means clustering algorithm(KFCM),which reconstructs a new objective function by means of a kernel function,thereby obtaining a better clustering effect.ECG is one of the main basis for the diagnosis of heart disease,evaluation of cardiac function.In this paper,the KFCM algorithm is used to analyze the ECG signal and the results obtained are satisfactory.And all ECG data were collected in the MIT-BIH arrhythmia database standards,and the original ECG data's baseline drift is calibrated and high frequency noise is removed.ECG data are generally high-dimensional data,which contains much of the redundant information,is not conducive to clustering data.On the one hand,data preprocessing can be performed through the method of correlation analysis;on the other hand,in order to improve the learning efficiency,there are more studies on dimension reduction,and the feature values reflecting the original ECG data can be extracted.Manifold learning is one of the main non-linear data dimensionality reduction,and we use the manifold learning's linear embedding algorithm(LLE)to feed pretreatment of ECG data dimensionality reduction.In order to compare the effect of dimensionality reduction,we also do data dimensionality reduction of the classical principal component analysis(PCA),and then we use the KFCM clustering algorithm and FCM clustering algorithm for the results of two methods of ECG data dimensionality reduction,and the clustering effect are compared and analyzed.From the experimental results,LLE algorithm for the extraction of ECG feature has not a better effect for the PCA algorithm,and since the impact of the selection of kernel function and its parameters of KFCM algorithm,the clustering effect is not stable and excellent compared with FCM algorithm.However,the computational complexity of KFCM is much smaller than that of FCM,and the clustering results can be obtained more quickly.