Research On Multi-classifier Integration Method For Disease Diagnosis

Studying a disease diagnosis model based on the integration of multiple classifiers is of great significance for improving the accuracy and timeliness of disease diagnosis and reducing the burden on medical personnel.For the numerical results of medical detection,machine learning methods can be actively studied and widely used in various medical diagnostic systems.Disease diagnosis is a complex decision-making process that can be solved using a classification method in machine learning.However,for unknown data,it is difficult to achieve high generalization ability if a single classification method is used for prediction.Therefore,in practical applications,multiple classifier combinations and optimization need to be considered comprehensively.Multi-classifier integration can significantly improve the generalization ability of a learning system,and thus has received extensive attention in the machine learning community.Based on the research of disease diagnosis methods and integration of multiple classifiers,this dissertation focuses on the multi-classifier integrated diagnosis model and optimization based on Support Vector Machine(SVM)as the base classifier.The main research contents and work results are as follows:(1)The traditional machine-learning-based disease diagnosis system was studied.Considering the unbalanced,redundant and high-dimensional characteristics of disease data,the data preprocessing,dimension reduction and unbalanced data processing methods were studied.Based on the typical SVM and KNN methods,a single classifier disease diagnosis model was established.Experiments were performed on diabetes and breast cancer datasets.The results showed that the SVM diagnosis method performed better predictive performance.(2)An in-depth analysis of multi-classifier integration methods was conducted.The Adaboost and Bagging algorithms were mainly studied.Adaboost and Bagging methods were used to change the distribution of samples respectively.A group of dissimilar single classifiers was obtained,which was based on Adaboost_SVM and Bagging_SVM.Integrated diagnostic model.The experimental results show that the multi-classifier integrated model performs better accuracy and stability in disease diagnosis than the single classifier model.(3)Based on the analysis of the defects of multi-classifier integration and selective integration methods,the base classifier diversity research was carried out.Attempts to study the selection criterion of the base classifier from the perspective of the balance between diversity and accuracy,and put forward a new WDA(Weighted Diversity and Accuracy,WDA),using genetic algorithm(GA)to design WDA and GA The multiple classifiers selectively integrate the diagnostic model and carry out relevant experiments on the disease data set,which provides a reference for effectively improving the diagnostic capabilities of multiple classifier integration systems.