The Prediction And Typing Of LGA Based On Machine Learning

Large for gestational age(LGA)indicates infants with gestational weight greater than the 90 th percentile of average gestational weight.LGA are prone to a variety of complications,and mothers are prone to all kinds of birth injuries.It is of great significance to establish a prediction model for LGA for early diagnosis and intervention.The research on classification of LGA conforms to the concept of precision medicine and is conducive to improving the benefits of diagnosis and treatment of LGA.In this study,data records of fetuses over the gestational age of 24-42 weeks born from 2010 to 2013 were selected as research objects,and machine learning technology was used to establish the prediction and subtyping model of LGA.The research on prediction and subtyping model of LGA is mainly divided into three parts: data processing,disease prediction and disease subtyping.The first step is to solve the problems caused by the originality and authenticity of sample information collection by applying data cleaning,data integration,data specification,and data transformation.The second step,sparse logistic regression model was used to classify disease and to obtain relevant features.Using Gradient boosting decision tree model to mine the nonlinear relationship between features and the classification results.The weak supervised learning was used to supply and generalize the prediction model of LGA by using data whose label was uncertain.The third step is to use positive samples and cluster method to subtype the LGA.The best result including recall rate,accuracy rate and area under the curve of classifier were 0.82,0.965 and 0.89,respectively.It was found that body mass index,smoking(passive smoking),life and work stress,and alcohol consumption were correlated with the incidence of LGA.The fetuses were less likely to be sick when indicators such as optimal reproductive age and hemoglobin of their parents were normal.There were two subtypes of large for gestational age infants,one of which was related to physical indicators such as male and female creatinine.And the other of which was related to social indicators such as occupation and education.In this paper,the relevant theoretical knowledge and real data in the field of prediction of LGA are used,and the machine learning method is applied to obtain a predictive classifier with good effects.The characteristics related to the occurrence of LGA are discovered,and two types of fuzzy subtypes are obtained.This is helpful for doctors to diagnose LGA,and the model has certain interpretability,thus achieving the goal of auxiliary clinic.