Design And Implementation Of Asthma Diagnosis Model Based On RBF-BP Neural Network

Recently,the impact of asthma on people’s health is increasing seriously.Among them,delay in diagnosis is an important factor affecting the cure rate of asthma patients.However,the existing diagnosis methods of asthma have disadvantages such as misdiagnosis,missed diagnosis,and high diagnosis cost.Therefore,at present,one of the hot topic in the field of asthma is how to diagnose asthma by using machine learning technology.The purpose of this study is to construct a model of asthma diagnosis with machine learning technology,which using the blood routine data of normal people and asthma patients.Firstly,based on relative weight and attribute reduction,we developed an approach of feature selection to deal with the data about blood routine.Next,we also develop a diagnostic model on the basis of RBF-BP hybrid structure neural network.Finally,an auxiliary diagnosis system for asthma is designed and realized.The following are specific work:(1)In order to solve the problem that existing feature selection methods ignore the potential relevance among data and put over reliance on specific models,which makes it difficult to find the optimal feature subset,a feature selection method based on relative weight and attribute reduction is proposed.The main idea is: on the one hand,,a relative weight estimation method with penalty coefficient is proposed to calculate the weight of each feature based on the reconstructed data set.So as to keep the correlation among the features to the maximum extent.On the other hand,a dynamic discretization method based on relative weight is proposed and used to process the blood routine data.Then,to calculate the optimal feature sets,a method of attribute reduction is developed,which is based on information entropy and combine the idea of greedy algorithm.The experimental results show that our proposed feature selection method has good performance in each model of machine learning.Meanwhile,compared with other feature selection methods,our method still performs higher accuracy and strong stability.(2)In order to solve the problem that the global mapping of BP neural network leads to slow convergence speed,and difficulty for RBF neural network to obtain completely linear separable mapping space for complex issues,a neural network diagnosis model is proposed on the basis of RBF-BP hybrid structure.In our model,the output of the hidden layer of the original RBF network is processed and cascaded with the hidden layer of the BP network,so that the local linear separable relationship in the RBF network can be transferred to the BP network.On the one hand,in the RBF part,in order to solve the problem of poor clustering effect caused by the uneven distribution of original samples,a dynamic clustering method based on cluster quality is proposed to select RBF centers.On the other hand,in the BP part,due to the over fitting problem of BP neural network,a regularization method is proposed,which adopt regularization method of relative weight and sample distance.According to the experimental results,RBF-BP model performs higher accuracy and stronger stability compared with other classification models,and the classification accuracy rate of 98.03% is obtained on the asthma data set.(3)Based on the above research and discussion,as well as blood routine data,an auxiliary diagnosis system about asthma is designs and realized in this work.The core function of the system is the intelligent diagnosis of asthma,as well as the auxiliary function of managing various medical information,patient information,drug prescriptions,work progress,department dynamics and other related information.