Research on risk factor screening and Nomogram prediction model construction based on unbalanced data
ZHOU Xuechao
HAO Baobing
LIU Chengyou
Abstract：In order to solve and optimize the problem that the classification results were biased towards the majority class due to unbalanced data in the process of pattern recognition,we took University of California,Irvine(UCI)myocardial infarction dataset as the research object,constructed a Nomogram prediction model.Firstly,three imbalance-handling methods,including K-fold cross-sam-pling voting(K-CSV),synthetic minority over-sampling technique_norminal continuous(SMOTE_NC)and random undersampling(RUS)were used to combine with mutual information,support vector machine weights,Spearman correlation analysis and variance ex-pansion facto to remove multicollinearity features.Secondly,univariate and multivariate Logistic regression were used to screen for inde-pendent risk factors,and constructe the Nomogram prediction model.The results showed that the original imbalanced data,the area un-der the receiver operating characteristic curve(AUC)value and average precision(AP)value of the model was 0.85 and 0.64,re-spectively.After RUS processing,the AUC and AP was 0.87 and 0.86 respectively,and the type Ⅱ error rate was 11.54％.After pro-cessing with SMOTE_NC,the AUC and AP were 0.96,but the accuracy rate dropped to 79.89％and the type Ⅱ error rate increased to 29.73％.The AUC and AP values of K-CSV were both 0.90,and the type Ⅱ error rate was 10.53％.The results of Cox regression anal-ysis showed that the selected features were significantly correlated with the prognosis of patients(P<0.01),indicating that the estab-lished model has high reliability in survival risk prediction.
Keywords：NomogramImbalanced dataSynthetic minority over-sampling technique_nominal continuousK-fold cross-sam-pling votingVariance inflation factorLogistic regression
Publication Date：2025-08-30
Online Publishing Date：2026-09-11(First online date of this platform, not the publication date of the document)
Pages：11( 245-255 )