Research On Medical Data Classification Based On Machine Learning

With the advent of the era of big data,medical datas show a geometric growth,how to effectively deal with and analyze the data is a very important issue.Traditional medical data classification methods often consume a lot of resources,and the accuracy is not high and unstable.Metaheuristic algorithms are more and more used to solve the problem of classification model optimization because of their fast convergence,but most studies focus on a single algorithm.The thesis comprehensively considers the advantages and disadvantages of different metaheuristic algorithms,and finally makes use of the coding characteristics of genetic algorithm(GA)to apply it to the discrete problem of feature selection.Considering the advantages of quantum particle swarm optimization(QPSO)algorithm in optimizing continuous parameters and convergence speed,it is applied to the hyperparameters optimization of machine learning algorithm.A hybrid optimization algorithm(GA-QPSO)based on the combination of GA and QPSO algorithm is proposed to synchronously optimize the feature selection and hyperparameters search in the classification process,and input to the random forest(RF)or k-nearest neighbor(KNN)classifier at the same time.Experiments are carried out on the biomedical datasets published in UCI,and the conclusions are as follows:(1)The GA-QPSO hybrid algorithm proposed in the thesis can achieve better classification performance for RF or KNN classifiers.In terms of classification accuracy,the classification evaluation criteria of hybrid algorithm GA-QPSO,including precision,F1 score and Kappa coefficient,they are better than those optimized by GA or QPSO alone.(2)The experimental results show that designing the cosine nonlinear decreasing method on the control parameter a of QPSO algorithm can further improve the global search and local search ability of QPSO algorithm,reduce the limitation of local optimal solution in the later stage of QPSO algorithm,and improve the accuracy and efficiency of QPSO algorithm.The improved MQPSO algorithm and GA algorithm are combined again to synchronously optimize RF and KNN classifiers.The experimental results show that the improved GA-MQPSO algorithm has further improved in the main indicators,which once again proves the necessity and advantages of hybrid synchronous optimization of GA and QPSO algorithms,and can provide some help and reference in dealing with medical disease diagnosis problems.(3)This thesis also analyzes the significance of each index to the actual medical diagnosis in the experimental results of the GA-QPSO hybrid algorithm,and analyzes that for the RF and KNN classifiers used for training in the experiment,it is more suitable for training binary or multi-classification datasets,which provides some reference for further exploring the optimization direction of the hybrid algorithm.