多粒度的老年人体能分级方法

针对目前评价老年人体能的方法存在检测项目繁琐、费时费力等弊端，提出一种多粒度的老年人体能分级方法.该方法首先采用多元回归、随机森林和BP神经网络3种方法进行融合决策筛选出11维与步速相关的重要属性.然后根据人群步速特征的分布将人群划归3个不同的步速层级.最后对相邻层级分别建立logistic回归模型，采用融合判定的方法将人群划分为7个体能等级.经实验验证，不同体能等级间人群的各属性（除臀围外）及失能得分情况具有明显的统计学差异.该体能分级方法可以用于评估老年人体能状态，有利于干预指导方案的制定. 

A Multi-Granularity Stratification Method for the Elderly Physical Fitness

A multi-granularity stratification method for the elderly physical fitness was proposed to overcome the tedious test items and time-consuming problem. Firstly, eleven key features were selected by three methods-multiple regression, random forest and BP neural network. Secondly, the individuals were divided into three different walking speed grades according to their speed characteristics. Finally, two logistic regression models were trained with the adjacent level of speed grades, and the physical fitness problem was stratified into seven levels based on the ensemble classification method. The experiment results show the basic features and disability score of people in different physical fitness levels have obvious statistical difference. The method can be utilized to assess the elderly physical fitness, and be beneficial for formulating the procedures and guidance.