Improvement Of Soft Interval Support Vector Machine Algorithm And Application Research On Heart Sound Classification

In the era of big data,data mining technology has been widely developed and used.However,in the process of data mining,better solutions are still needed to accurately select feature data and improve the classification efficiency of feature data.In the classification algorithms of data mining,support vector machines have the advantages of simple structure,global optimization,and strong generalization ability.Over the years,with its robust mathematical foundation and theoretical support,a large number of researchers have widely applied support vector machines in various fields and continuously improved algorithms through various methods,resulting in significant classification results in specific applications.This article focuses on improving the classification efficiency of feature data,focusing on the classification performance of support vector machine methods,and proposes improved methods.The main work of this article can be summarized as follows:Firstly,this article carefully interprets the support vector machine theory and makes improvements based on its current shortcomings.Aiming at the problem that the classification process of soft interval support vector machine is affected by parameters,the grey wolf optimization algorithm(GWOA)and particle swarm optimization(PSOA)are proposed to replace the traditional grid search method to find the optimal parameter combination of SVM,and the GWOA-SVM and PSOA-SVM classification models are constructed;Then,in order to further enhance the classification performance of GWOA-SVM and PSOA-SVM,and reduce the generalization error of the classification model,corresponding solutions are proposed respectively: First,the improved gray wolf optimization algorithm(IGWOA)is used to replace GWOA to find the optimal parameter combination of SVM,and the IGWOA-SVM classifier model is constructed;Secondly,using the Stacking based ensemble approach,the PSOA-SVM classification model is used as a primary classifier to integrate other primary classifiers(Adaboost and RF)and secondary classifiers(LR),and an Adaboost+RF+(PSOA-SVM)+LR classifier model is constructed.Finally,based on a publicly available heart sound database,a binary heart sound dataset was obtained after denoising preprocessing,heart sound segmentation,and feature extraction of the heart sound signal,and experiments were conducted on heart sound classification.This article uses this heart sound dataset to train and generate 10 classifier models.Comparative experimental results show that the two classification models proposed in this article have shown excellent classification performance in specific classification experiments,with classification accuracy reaching 98.15% and 97.38%,respectively,and their specificity reaching 100%.Therefore,it can be proven that this study provides a more accurate classification model for the application of heart sound classification in the medical field,which has important practical significance.