Research On Neural Network Classification Method Of Medical Data Based On Meta-heuristic Algorithm

In the occasion of medical informatics,the accuracy of medical data classification plays a material role.With the unceasing blossom of artificial intelligence technology,disease classification and diagnosis in the healthcare industry has become a fashionable adhibition region of artificial intelligence,and machine learning-assisted decisionmaking systems are widely used to assist doctors in disease diagnosis.Therefore,it is necessary to study how to improve the classification accuracy of medical data.There are currently three main approaches to this classification problem.Machine learning based medical data classification methods are one of the most frequently used approaches in this field.On the side,due to the continuous development of heuristic algorithms in cybernetics theory,more and more researchers train hyperparameters of classifiers based on heuristic optimization algorithms to achieve classification tasks.Finally,the problem of medical data classification is solved by the method of feature selection.This thesis analyzes the research status of these three mainstream classification methods,and mainly carries out related research from the latter two classification methods.Based on heuristic algorithms,this thesis proposes four methods for solving the medical data classification tasks.(1)A multi-layer perceptron classification method based on probability distribution-driven bio-geographic optimization algorithm(PDBBO-MLP)is proposed to complete the medical data classification task.Eight different probability distributions are used to improve the migration process of the original biographical optimization algorithm,and eight algorithm variants are proposed.Based on these different variants,the classification models are constructed respectively,and the simulation verification is carried out on 6 groups of medical datasets.The simulation results show that the eight PDBBO-MLPs can significantly improve the classification accuracy.Among them,the biographical optimization algorithm based on exponential distribution has the highest classification accuracy.Finally,the proposed classification model is compared with other classification methods,which once again confirms the utility of the proposed method.(2)A long-short-term memory network classification method based on chaotic arithmetic optimization algorithm(CAOA-LSTM)is proposed to complete the task of diabetes data classification.Using two key parameters of the ten chaotic interference arithmetic optimization algorithms,the mathematical optimization acceleration parameter(MOA)and the mathematical optimization probability parameter(MOP),ten improved algorithms are proposed.Based on these algorithms,ten kinds of classification models are constructed by combining with long-short-term memory network(LSTM)respectively,which are verified by simulation on two sets of diabetes datasets.The simulation results show that 10 kinds of CAOA-LSTM can greatly improve the classification accuracy.Among them,the AOA based on Tent mapping has the highest accuracy.(3)A bidirectional long-short-term memory network classification method based on the trigonometric function pedigree arithmetic optimization algorithm(LTSAOABi-LSTM)is proposed to complete the diabetes classification task.Based on 4 types of trigonometric functions(regular trigonometric functions,inverse trigonometric functions,hyperbolic trigonometric functions,and inverse hyperbolic trigonometric functions),24 variable step size coefficients for Lorentzian trigonometric search(LTS)are proposed.It replaces the mathematical optimization probability parameter(MOP)in the AOA to reconstruct the position update of the algorithm.This series of algorithms are collectively referred to as trigonometric function pedigree AOAs.Through the comparison of simulation experiments,it is found that the optimization effect of the search coefficient based on the tangent family is the best.Combining four AOAs based on trigonometric functions of the tangent family with long-short-term memory network(LSTM)and bidirectional long-short-term memory network(Bi-LSTM)respectively.Eight classification models are constructed and validated by simulation on two diabetes datasets.The emulation outcome demonstrate that the classification grade of the BiLSTM-based classification model is commonly taller than that of the LSTM-based classification model.Among these classification algorithms,the AOA based on the tangent search step size has the highest classification accuracy.Finally,the proposed classification model is compared with other approaches,which once again confirms the effectiveness of the proposed method.(4)This thesis designs a medical data feature selection method based on the logical binary sine and cosine arithmetic optimization algorithm(LBSCAOA-KNN).Based on the classic S-type and V-type transformation functions,the AOA on the successive province is transformed into a binary arithmetic optimization algorithm(BAOA)on the discrete domain.The logic operation strategy and the sine-cosine interference strategy are applied to it,and an improved version of the logic binary sine-cosine arithmetic optimization algorithm(LBSCAOA)is proposed.Based on the algorithm,the simulation verification is carried out on 10 groups of UCI medical data sets and 10 groups of UCI ordinary data sets.The simulation results show that the use of LBSCAOA-KNN can significantly improve the classification accuracy of medical data.