比例优势boosting算法在高维有序多分类数据分析中的应用

摘要

目的 探讨比例优势boosting算法在高维组学多分类有序数据中变量筛选和分类预测的应用.方法 通过模拟实验和实例比较比例优势boosting算法和其他常用的多分类boosting算法在变量筛选和分类效果中的差异.结果 模拟实验表明,比例优势boosting算法的变量筛选效果,尤其在小样本情况下要明显优于其他方式,分类效果略优于其他方式;实例数据分析结果表明,比例优势boosting变量筛选效果要优于其他方式,在分类效果上略低于随机梯度boosting, 但优于其他boosting方式.结论 比例优势boosting算法适用于高维有序多分类数据,具有实用价值.%Objective To explore the application of proportional odds boosting(P/O Boosting) model in variable selection and classification in the case of ordinal multiclass high dimensional data. Methods We used simulated experiment and actual data to compare the variable selection and classification of P/O Boosting and other common multiclass boosting algorithms. Results Simulation experiment suggested that P/O Boosting usually performs better than other methods, especially for small samples. Analysis results of actual data suggested that P/O Boosting outperforms than others in variable selection and except SG-BT in classification. Conclusion P/O Boosting is applicable to ordinal multiclass high-dimensional omics data and possessed practical value.