| Schizophrenia is a complex multifactorial disease that has a certain impact on society and family.Research based on Single Nucleotide Polymorphism(SNP)is an important subject in the field of biomedicine.With the development of data mining technology,many researchers use machine learning methods to select feature of SNP sites and build diagnostic models for schizophrenia.In this thesis,SNP sites in schizophrenia are studied.Firstly,the feature selection of SNP data is carried out based on the improved K-Center algorithm and the improved Particle Swarm Optimization algorithm,and then the schizophrenia diagnosis model based on Adaboost algorithm is established.The specific work is as follows:(1)Aiming at the problems of linkage disequilibrium between SNP sites,a new algorithm,K-MSU,is proposed to cluster SNP data.K-MSU introduces the concept of symmetric uncertainty in the K-Center algorithm and defines a new distance measurement formula,which effectively solves the problem that the original distance measurement formula cannot mine the correlation between SNP sites.Aiming at the problem of randomly selecting the initial cluster centers in the K-Center algorithm,the information gain is introduced into the density function to measure the contribution of each information SNP site,and the cluster centers are selected selectively.Experiments show that K-MSU algorithm has better clustering effect and prediction accuracy compared with other methods,and its classification accuracy is increased by 1.43% and2.38% on average in the datasets Dataset1 and Dataset2.(2)In the process of feature selection,the Particle Swarm Optimization algorithm will make the particles fall into the local optimum when optimizing,which results in poor global search results.In response to the above problems,this thesis proposes a new fitness calculation method and inertial weight update method.Firstly,the accuracy of the information SNP subset for non-information SNP reconstruction is introduced into the fitness function of the Particle Swarm Optimization algorithm to search for the best SNP features in the current region.Then the redundancy of the information SNP subset is introduced into the inertia weight,and the inertia weight is dynamically adjusted to enhance the global search capability.Through experiments,it is found that compared with other feature selection methods,the improved particle swarm algorithm has a better effect on the reconstruction of non-information SNP,and its classification accuracy is increased by 5.33% and 5.03% on the datasets Dataset1 and Dataset2 on average.(3)In medical diagnosis,the misclassification costs of diagnosing a patient as a healthy person and a healthy person as a patient are different,and the impacts are also different.Therefore,a cost-sensitive Adaboost algorithm is proposed,which uses entropy value to calculate the weight of misclassification cost,and introduces it into the update of sample weight of Adaboost algorithm,adjusts the weight of each base classifier,and integrates the final classifier to diagnose patients with schizophrenia.Experiments have found that although the accuracy rate of the schizophrenia diagnosis model has no obvious advantage,the error rate of the misclassification cost of the data sets Dataset1 and Dataset2 has decreased significantly,down by 5.36% and 4.72%,respectively,which proves that the model is more suitable for schizophrenia Diagnosis. |
