Construction Of Prediction Model Of Eating Disorder Tendencies Of Nursing College Students Based On Machine Learning

Objectives:The study aimed to evaluate the prevalence of Eating Disorder(ED)tendencies among Chinese nursing college students,and to identify the optimal Machine Learning(ML)prediction model of ED tendencies by constructing and comparing multiple ML prediction models.Meanwhile,this study intended to analyze the influential factors of ED tendencies in nursing college students based on the theoretical framework of ED risk and protective factors,and to provide guidance for nursing educators to early identify and prevent high ED risk in nursing college students.Methods:This was a cross-sectional study conducted between March 2022 and June 2022.A total of 1709 nursing college students from several universities in China were enrolled Online by a convenient sampling method to conduct a questionnaire study.The research tool was constructed based on the theoretical framework of ED risk and protective factors,including questionnaires and scales from six aspects:basic demographic,physiological,psychological,behavioral,family,and social factors.In this study,R(version 4.0.5)and Python(version 3.7.9)were used for statistical analysis of data.The steps of statistical analysis include four parts:feature variables screening,prediction model construction,prediction model evaluation and comparison,and results visualization.In this study,the feature variables screening was conducted to construct dataset of prediction models.Furthermore,the original data set was divided into a training set and a test set in a ratio of 7:3.The training set was used to construct the prediction models;the test set was used to evaluate and compare prediction models.In the feature variables screening part,this study constructed two feature datasets from the perspectives of predictability and practicality,respectively:the former was named as Data1 in which the variables were screened using univariate analysis with Bonferroni correction P value,and the latter was named as Data2 in which the variables were further screened by two-way stepwise Logistic regression.Three ML algorithms including Logistic regression,random forest,and BP artificial neural network were used to construct the prediction model,and the models’ parameters were optimized by 10-fold cross-validation method and grid search method.In this study,AUC and Average Precision(AP)were calculated,and ROC curve,PR curve and DET curve were drawn to evaluate and compare the prediction performance of the prediction models to select the optimal prediction model with advantages in predictability or practicability aspect.Finally,this study visualized the feature importance ranking results and drawn a nomogram.Results:1.Among 1709 nursing college students recruited in this study,the prevalence of ED tendencies was 39.4%.2.In this study,the good prediction performance was founded in the Logistic Regression Model-1(LR-1),Random Forest model-1(RF-1)and BP Artificial Neural Network Model-1(BP ANN-1)which was constructed based on Data1.The AUC and AP of LR-1 were 0.789(0.778～0.798)and 0.695(0.681～0.708)respectively;RF-1 were 0.816(0.801-0.831)and 0.729(0.703～0.752);and BP ANN-1 were 0.782(0.767～0.794)and 0.691(0.672～0.708).RF-1 had the best predictive performance,while LR-1 and BP ANN-1 had similar predictive performance.3.In this study,the good prediction performance was also founded in the Logistic Regression Model-2(LR-2),Random Forest Model-2(RF-2)and BP Artificial Neural Network Model-2(BP ANN-2)which was constructed based on Data2.The AUC and AP of LR-2 were 0.793(0.783～0.801)and 0.715(0.702～0.724)respectively;RF-2 were 0.804(0.789-0.818)and 0.711(0.682～0.737);and BP ANN-2 were 0.793(0.783～0.800)and 0.711(0.699～0.721).The predictive performance of RF-2 was the best,while that of LR-2 and BP ANN-2 was similar.4.Based on the theoretical framework of ED risk and protective factors,the results of LR and nomogram showed that the greatest risk factor of ED tendencies of nursing college students was rumination related to eating,body shape and weight,followed by history of childhood abuse,low body satisfaction,perceived stress,depression,overweight or obesity,low frequency of vegetable consumption,insomnia,emotional eating,and religious belief.5.The feature importance ranking results of the LR and RF are mostly consistent,both of which believed that rumination related to eating,body shape and weight,and depression were important factors influencing ED tendencies of nursing college students.However,the RF results showed that childhood abuse was not an important factor influencing ED tendencies of nursing college students.Conclusion:The prevalence rate of ED tendencies in nursing college students is high in China,so nursing educators need to pay attention to the early prevention and intervention of ED risk in nursing college students.In the prediction model of ED tendencies,the prediction model constructed by random forest algorithm has high prediction and practical value.Specifically,from the perspective of predictability,RF-1 is the best,followed by RF-2.From the perspective of practicability,RF-2 is the best,followed by LR-2.In order to effectively prevent the risk of ED in nursing college students,nursing educators should first pay attention to the intervention of rumination related to ED in nursing college students,such as taking short-term mindfulness training.Secondly,nursing educators should take attention to the protection of focus groups,such as the nursing college students with history of childhood abuse,religious belief,or overweight and obesity,give them regular psychological counseling and improve their social support.In addition,nursing educators can regularly monitor students for insomnia,depression and stress and implement relevant interventions.Finally,nursing educators should attach importance to ED education and urge nursing college students to correct unhealthy aesthetic concepts and reduce abnormal eating behaviors.