Prediction of Alzheimer's Disease Progression based on Resampling and Integrated Classifiers
10.11783/j.issn.1002-3674.2025.05.011
- VernacularTitle:基于重采样和集成分类器的阿尔茨海默病进展的预测研究
- Author:
Weiqiang SU
1
;
Shuting DING
1
;
Mengyao GAO
1
Author Information
1. 山东第二医科大学公共卫生学院(261053)
- Publication Type:Journal Article
- Keywords:
Machine learning;
Resampling;
Classifier;
Alzheimer's disease
- From:
Chinese Journal of Health Statistics
2025;42(5):699-704,712
- CountryChina
- Language:Chinese
-
Abstract:
Objective In this study,a risk prediction model for Alzheimer's disease(AD)was constructed based on Alzheimer's disease neuroimaging initiative(ADNI)database to provide a theoretical basis for a better understanding of AD,as well as to provide a reference basis for the rational allocation of health resources and the development of preventive and control strategies.Methods The ADNIMERGE,UPENNBIOMK,ADNINIGHTINGALELONG and DXSUM datasets from the ADNI database were downloaded and random forest algorithm was used for the missing values imputation.Variable screening was performed by LASSO regression.Borderline-1 SMOTE resampling was used to adjusted for intergroup balance,and the variables were incorporated into base classifiers including XGBoost,random forest,bagging,AdaBoost,and support vector machine.Enhanced integrated classifiers were then constructed based on voting and Stacking strategies.Youden index,area under curve(AUC),F-score,G-mean,accuracy,Matthews correlation coefficient(MCC)and Kappa on the validation set were used to evaluate and compare the model efficacy.Results The performance of the classifiers based on balanced data were improved for both base and enhanced integrated classifiers.The performance of stacking and voting enhanced integrated classifiers constructed based on advantageous base classifiers have better performance compared to the base classifiers.After data balancing,the XGBoost performed better in the base classifiers(AUC:0.9090,accuracy:0.9091)and voting algorithm performs better in enhanced integrated classifiers(AUC:0.9178,accuracy:0.9179).Conclusion After Borderline-1 SMOTE resampling adjusted,the performance of both base classifiers and the enhanced integrated classifiers were all improved.For balanced data,XGBoost classifier and the voting enhanced integrated classifier can effectively assist in clinical prediction of Alzheimer's disease progression.
