Research On Multi-classification Algorithm Based On SVM And Adaboost

The accelerated pace of life has caused many problems in people's sleep health and emotional processing.Therefore,sleep stages and emotional recognition have become one of the hotspots concerned by researchers nowadays.Sleep stage and emotional recognition can be used as adjuvant therapy for insomnia,depression,anxiety and other diseases.A large number of studies have shown that the existing algorithms have the following shortcomings: 1)Classical classification algorithms are mostly based on binary classification,which requires multiple binary classification processing when doing research on many kinds of problems,with high time complexity;2)Classical classification algorithms have relatively low classification accuracy,which cannot meet the requirements of sleep stages and emotional recognition research;3)Most of the common classical algorithms are based on binary classification.It can only solve the problem of binary classification,and the algorithm structure needs to be improved.In view of the above problems,this paper designs and proposes two multi-classification algorithms for sleep staging and emotion recognition,which solves the problem that traditional algorithms cannot be directly applied to multi-classification research.This paper improves the structure of the algorithm,which reduces the complexity of the algorithm,and improves the classification accuracy of the algorithm.The main contents of this paper are as follows:(1)This dissertati combines decision tree with least squares support vector machine(LSSVM),and proposes a decision tree and least squares support vector machine(DLSVM).LSSVM instead of leaf nodes in decision tree can be directly applied to the study of multi-class problems.DLSVM algorithm has the following advantages: on the one hand,it can reduce interference factors and gradually decompose complex multi-classification problems into simple binary classification problems to improve the accuracy of classification algorithm;On the other hand,the LSSVM classifier is embedded in the root and middle nodes of the decision tree.With the help of the structural advantages of the decision tree algorithm and the two-class attributes of the classifier,an algorithm that can directly carry out multi-classification is constructed.When entering the next level of algorithm,the labeled data does not participate in the next level of classification algorithm.Therefore,after one level of iteration,the data complexity of the algorithm decreases,the time complexity of the algorithm is reduced,and the classification accuracy of the algorithm is improved.(2)This paper improves the Adaboost algorithm,and proposes a multi-classification algorithm based on hierarchical structure and Adaboost(HAdaboost)to deal with multi-class problems.The HAdaboost algorithm can adjust the number of layers of the algorithm adaptivelyaccording to the number of classifications of the data.After the iteration of the one-layer algorithm,a classification mark can be obtained.In this paper,the above algorithm is applied to sleep automatic staging and emotion recognition,and the algorithms achieve good classification results.