Optimization Of SVM Classification Model For Defecation Dysfunction Based On Particle Swarm Optimization

Many disorders are related to abnormal intestinal motility,among which chronic constipation is a common one.The chronic constipation can cause pain of the human body and it’s incidence is high,which thereby greatly affects the life quality of the patients.Therefore,it is important to study the optimal management of bowel motility,early diagnosis and prevention.However,there is no effective method to distinguish patients with defecation dysfunction from healthy subjects.This study aims to extract efficient features that characterize colonic motility based on the physiological parameters of the anorectal system,which are collected by the Fecobionics bionics equipment in the natural defecation process.The differences of rectal motility among different types of subjects are identified,and an effective automatic model for classifying rectal motility is established.Based on healthy,constipation and fecal incontinence subjects,features are extracted from multiple perspectives and dimensions,which is different from those of the existing rectal motility analysis method.Then a subset of features is selected according to the correlation among different features.Finally,the SVM classification model for three classifications optimized by particle swarm optimization is established.To evaluate the performance of the proposed method,the proposed model is compared with KNN,Naive Bayes,and SVM classical classification models based on precision,recall and accuracy.The results show that the proposed model is effective with high prediction accuracy,precision and recall.The main works and results of this study are as follows:(1)Based on Fecobionics equipment,the physiological parameters of anorectal system in the natural defecation process were collected.After denoise the data are analyzed according to the relavant knowledge of intestinal system motility.Meanwhile,the informatics analysis method is adopted to characterize different features.The features mainly include: loop area,force index,information entropy,multi-scale entropy,and Hurst exponent of fractal features.In addition,statistical features are used as auxiliary features.These features are of great significance for the diagnosis of defecation dysfunction.Finally,the optimal feature subset is obtained by feature selection.(2)Several classification prediction methods are analyzed,and various algorithms to predict the types of defecation dysfunction are implemented,including KNN algorithm,Na(?)ve Beyes algorithm and SVM algorithm.KNN and SVM classification algorithms build a classification model through training samples,and then test samples are use to evaluate the model.Na(?)ve Beyes is based on probability and does not require a high number of samples.The classification accuracy of these classification prediction methods are compared with the test samples.It is found that both of the SVM model and the Na(?)ve Bayes model perform well with the accuracies of 0.7778 and 0.7776 respectively.(3)Based on particle swarm optimization,a SVM multi-classification model is proposed.For the three-classification problem in this study,a multi-classification SVM model suitable for the data is introduced.The particle swarm algorithm is used to find the optimal parameters so that the model can have better classification effect.Through multiple experimental tests,it is found that when the kernel function adopts RBF with a penalty coefficient C of 102.938 and a kernel parameter g of 0.02,the classification accuracy of the cross-validation of the model can reach a highest value of 0.8318.The results validates that this method can improve the classification accuracy of the model.