Research On The Gestational Diabetes Mellitus On Data Mining

With the development of computer technology,many computer information management systems have been heavily used in medical institutions,which has made the rapid growth of medical information of medical institutions.In addition,the form of medical information is varied,including text,tables,images,audio,etc.The medical information resources play an important role in decision-making management,diagnosis and treatment and scientific research of medical institutions.One of the problems that must be solved is how to extract valuable information and models from these complex information.Based on the complexity and particularity of medical information,data mining,as a kind of data processing,provides a method to realize the effective utilization of medical information resources.This thesis applies data mining technology in the study of gestational diabetes mellitus.First of all,analyze influencing factors of gestational diabetes mellitus using association rules to identify high risk single factor,including PCOS medical history,family diabetes history,ectopic pregnancy history,early pregnancy age,marriage age,the gestational age and whether or not to be multiple births and ever macrosomia birth.At the same time,some high risk factors have discovered,such as:when multiple births and the history of menstrual cycle disorder exists,the risk of pregnant women with gestational diabetes mellitus is as high as 100%;The risk of individual illness is less than combinations including ectopic pregnancy history and multiple births,menstrual cycle disorder history and PCOS history,menstrual cycle disorder history and family diabetes history,etc.Secondly,forecast the risk of gestational diabetes mellitus using classification algorithm.By extrapolation forecast and cross validation,compare four commonly used classification model——Logistic regression,decision tree,support vector machine(SVM)and BP neural network model.Finally,select the BP neural network to establish the prediction model of gestational diabetes mellitus according to the model prediction effect and stability.In establishing the model,training data is balanced by SMOTE.Compared with the model which is built by original unbalanced data,the model which is built by balanced data predicted results have been greatly improved,especially forecast for gestational diabetes mellitus with 100%accuracy for the top 20%,thus,maternal with high risk of gestational diabetes mellitus is well identified.