Research On Training Method Of Support Vector Machine And Its Application In Disease Diagnosis

As a major threat to human health,diseases are not only related to the unhappiness of each family,but also to the stability and development of the entire society.Accurate disease diagnosis results play a crucial role in its prevention and treatment.Disease diagnosis is a complex decision-making process.The traditional disease diagnosis depends mainly on the experience and professional level of the doctor.With the widespread use of artificial intelligence technology,more and more new technologies are being applied to disease diagnosis to help doctors make better decisions.As a commonly used classification model,SVM(Support Vector Machine)has good robustness and adaptability when used to classify the small samples and high-dimensional data,and it has achieved good results in the diagnosis of diseases.However,in the support vector machine training process,there are a large number of constraints that are meaningless for determining the final hyperplane.In this thesis,the training method of support vector machine is researched,and an algorithm,SVMPM,is proposed.An ensemble learning method based on the trained SVM is applied to the disease diagnosis to help doctors get more accurate diagnosis results.In this thesis,the main research contents can be summarized as follows:1.Aiming at the problem of too many constraints in the training process of support vector machine,a constraint reduction strategy is proposed.Firstly,the QuickHull algorithm is used to find boundary points of training samples.Secondly,the distance between all sample points belonging to different classes are calculated and sorted.Finally,the reduction of meaningless constraints is completed by deleting the sample points whose distance is higher than the preset threshold.The experimental results show that the support vector machine model with constraint reduction strategy can maintain the same classification performance as the classic support vector machine.2.In this thesis,a penalty method based SVM training method is proposed,and an ensemble model which combining the improved support vector machine is established for disease diagnosis.Firstly,the constraint reduction strategy proposed is used to reduce the training samples.Secondly,the penalty function method is used to transform the SVM training problem into an unconstrained optimization problem and the gradient descent algorithm is used to solve the unconstrained optimization problem.Finally,an ensemble model for disease diagnosis is established by ensembling the improved support vector machine and two other basic classifiers.The experimental results show that the ensemble model has better classification effect than the classical SVM and other classifiers.