Research On Diabetes Prediction Model Based On SVM

In recent years,with the continuous progress of science and technology and the development of big data technology,our country’s comprehensive digital construction has been promoted.The application of machine learning algorithms to the medical field has greatly promoted the development of online medical care,which has become a research hotspot today.In the medical field,diabetes is one of the three major chronic diseases.Its treatment cycle is long and complex,and there is no immediate treatment.Accompanied by the aggravation of the disease,a series of complications will occur,which directly affects the cardiovascular and cerebrovascular organs of the patient.Therefore,we should analyze the complex big data to extract useful feature information,then use machine learning to establish a diabetes classification prediction model,meanwhile,strengthen the screening and self-testing of early diabetes.Diagnosing diabetic patients in time can effectively control the further aggravation of the patient’s condition.In this way,the subsequent harm caused by diabetes to patients can be controlled to the greatest extent.Machine learning algorithms can effectively establish mathematical connections from numerous features to obtain predictive models for diabetes.However,the existing algorithms for diabetes prediction have the following limitations: 1)The traditional machine learning algorithm trains faster but the prediction accuracy is lower;2)The deep learning model has a complex structure,which requires a large amount of data and high computing resources.Also,it takes a lot of time to train,which makes it difficult to deploy in real applications.In view of the above problems,this paper applies the Support Vector Machine（SVM）in the traditional machine learning algorithm to the prediction of diabetes,and proposes a diabetes prediction model with the optimization goal of improving the prediction accuracy of SVM.The optimization and improvement of the diabetes prediction model is divided into three parts: optimizing feature processing,proposing a new SVM kernel and optimizing with swarm intelligence algorithm.In the first part,in order to solve the problem that the distribution of the origin sample set is scattered and difficult to divide,this paper trains the sample features with the loss function Focal Loss and Center Loss,so as to enlarge the distance between different categories and compact the internal distance of the same category.In the second part,in order to solve the problem of sample distance dispersion in highdimensional space after the samples are mapped,this paper uses the metric functions LMNN,NCA and MLKR to construct a new SVM kernel,and proposes a new SVM kernel based on the metric function as a new kernel of the support vector machine（Metric Support Vector Machine,MSVM）,the proposed new kernel function can compact the distance between the same kind,which is more conducive to dividing the classification hyperplane.In the third part,in order to solve the limitation of the small search range of the Sparrow Search Algorithm（SSA）and avoid falling into the local optimal solution,the SSA is optimized by combining the elite opposition strategy and the firefly disturbance strategy.The optimized algorithm is EOF-SSA,and its initial search range becomes larger and can jump out of the local optimal solution,which is used to optimize the adjustment factor of the new kernel of MSVM proposed in this paper.The experimental results show that on the Pima diabetes data set,the EOF-SSAMSVM diabetes prediction model proposed in this paper pays less time than the deep neural network model,and the prediction accuracy is greater.The final experimental prediction reached 86.3%,while the prediction accuracy of the traditional SVM model was 77%,and the accuracy of the model in this paper was improved by 9.3%.Therefore,this paper achieves an improvement in the accuracy of diabetes prediction problem.