| Depression is a complex and heterogeneous psychiatric disorder,and the current detection of depression is still very difficult,mainly because its pathogenesis has not been fully elucidated,and the lack of objective clinical diagnostic criteria for depression detection.This thesis studies the application of machine learning and deep learning in depression risk detection methods,in order to provide reference for domestic research in this field.Starting from the data of hospital physical examination results,this thesis studies the basis of processing and analysis of physical examination data information,and uses machine learning and deep learning in artificial intelligence for depression risk detection,and the main research content is as follows:Firstly,aiming at the problem of data loss,a missing value prediction method based on random forest is proposed,and its effect is relatively good.The important features were screened by the variance filtering method of feature engineering,and the importance and correlation of the selected features were analyzed,and the important feature factors and the correlation with depression were mined,which laid a data foundation for improving the accuracy of model prediction.Secondly,a machine learning-based depression risk detection model was studied.A variety of machine learning algorithms were constructed as comparative experiments.On this basis,a method of Vowing and Stacking model fusion based on comparative experiment is proposed.Using grid search optimization parameters,using ten-fold crossvalidation for training,using performance indicators for evaluation,the accuracy,AUC value,F1 value,recall rate and accuracy of different models were obtained,and the comparison and judgment were made,and finally the model with the best prediction effect,namely the Stacking model,with an accuracy rate of 0.934,was selected to provide reference for the prediction and diagnosis of depression.Thirdly,a depression risk detection model based on deep learning was studied to further detect depression.A variety of network models were constructed as comparative experiments.On this basis,an SSA-LSTM model coupled with long short-term memory network and sparrow algorithm and CNN-LSTM-Attention with attention mechanism are proposed,and performance indicators are used to evaluate.The results show that the proposed model has better performance than the basic network model or even the machine learning algorithm.The CNN＿LSTM＿Attention accuracy rate reached 0.945.This shows that the model established in this thesis improves the accuracy and stability of the overall classification of depression to a certain extent,and provides new methods and ideas for the modeling and analysis of depression risk prediction.Finally,the work of this thesis is summarized,and the shortcomings of this thesis and the future research direction are clarified. |
