Diagnosis And Prediction Of Diabetes During Pregnancy Based On Integrated Learning

Diabetes Mellitus is one of the major health challenges all around the world.Gestation diabetes mellitus(GDM)is defined as glucose intolerance detected during pregnancy.The prevalence of GDM is increasing at a fast pace,which jeopardizes the Maternal-Child Health and influences the social development to a great extent.The healthcare community is increasingly gaining interest to the prediction and prevention of GDM.The motive of this study is to design a model which can predict the suffering of GDM with maximum accuracy.Experiments are performed on groups of 1000 pregnant women’s clinical and genic data which comes from a data competition launched by Qing Wu Tong Gene Health Technology Company and Alibaba.Feature selection of Wrappers and Embedded has been done in this research,and the selected features identified by Wrappers perform better than the others.Then we used four classification algorithms namely Decision Tree,Logistic Regression,Random Forest and Extreme Gradient Boost to detect GDM.Results show Xgboost performs better than other algorithms.Then we integrate all the results from these previously mentioned models by four different ways named averaging,the PCA weight method,the entropy weight method and stacking algorithms.Overall the performance of stacking is better than other methods.These results are verified using Receive Operating Characteristic(ROC)curves,F1 score and recall score.We also find that high VAR00007,age,obesity,multiple pregnancies and high diastolic blood pressure are risk factors for GDM.