| With the improvement of people’s living standards,the change of dietary habits and the implementation of the two-child policy,the proportion of older mothers is increasing,which further leads to the rise of the incidence of gestational diabetes in China.Meanwhile,the incidence of gestational diabetes is also showing a rapid growth trend worldwide.Gestational diabetes can lead to adverse pregnancy outcome,make the pregnant women and fetal health be affected,now still no effective treatment can be treated for gestational diabetes,therefore,it is of great practical significance to screen and intervene early in the population prone to gestational diabetes,which can prevent the occurrence and development of the disease.In recent years,with the development of intelligent medicine,deep learning and other technologies,how to make effective analysis of valuable medical data has become a research hotspot.Researchers have applied deep learning techniques to the medical field to predict,intervene and manage disease,and have made significant progress.Many scholars generally use traditional machine learning algorithms to predict diseases.According to the current development status of deep learning in disease prediction,this paper uses deep learning technology to build a prediction model of gestational diabetes mellitus through deep neural network,and carries out research on gestational diabetes mellitus.The specific research work of this paper is as follows:(1)Data preprocessing was carried out to solve the problems such as incomplete and abnormal data from the GDM data set of Tianchi Precision Medical Contest,so as to provide a smooth and balanced data set for subsequent feature selection,this preprocessing process also provides some reference value for data processing of other diseases.(2)Aiming at the problem that the feature dimension of data set is high and there are a lot of redundant features which affect the accuracy of prediction model,XGBoost algorithm and IV value were used for mixed feature selection,and dimensionality reduction was performed on the data set,12 feature variables with high feature weight were selected from the preprocessed 73 features,which were combined into a new feature subset for the input of the subsequent prediction model,and comparative experiments were designed to prove that feature selection improved the model performance.(3)Aiming at the shortcomings of the traditional DNN network model,a prediction model based on batch normalization was proposed.The model introduced batch normalization(BN)layer on the basis of DNN,which effectively solved the problems of gradient vanishing and local optimal solution,at the same time,in order to obtain the best model effect,the multi-layer grid search algorithm,L2 regularization,Adam optimization algorithm,dropout layer,10 fold cross validation and other methods were used to optimize the model.In order to verify the effectiveness of the model,the model is compared with other algorithms,and the experimental results show that the model is superior in accuracy,sensitivity,specificity,precision and so on. |
