| 摘 要: 研究比较多重线性回归模型､径向基(RBF) 神经网络模型､反向传播(BP) 神经网络模型在矽肺
发病工龄预测中的适用性｡方法 以2006—2015 年报告的河北省壹期矽肺病例为研究对象建立数据库, 并将实际发病工龄与三种模型预测的发病工龄进行配对秩和检验, 计算预测值的平均相对误差和平均绝对误差｡结果 共获得壹期矽肺2 294 例, 经秩和检验, RBF 神经网络模型的预测值与发病工龄的差异有统计学意义(P<0.05), 其余两种模型差异均无统计学意义(P>0 05), 其中BP 神经网络模型的预测精度最高; 开始接尘年代在矽肺发病预测中占得权重最大｡
结论 应根据数据特点及分析需要选择适宜模型, 在矽肺发病工龄预测中BP 神经网络模型优于多重线性回归模型｡ |
| 关键词: 壹期矽肺 接尘工龄 多重线性回归模型 ＲＢＦ 神经网络模型 ＢＰ 神经网络模型 |
|
中图分类号:
文献标识码:
|
| 基金项目: 河北省卫生计生委医学科学研究重点课题， 河北尘肺
病流行规律与防治对策研究（２０１３００８９） |
|
| Comparative study of three statistical models in prediction of working age for silicosis |
|
Zhao Junqin, Li Jianguo, Liu Huitian, Zhao Chunxiang
|
|
Hebei Provincial Center for Disease Prevention and Control, Shijiazhuang 050021, China
|
| Abstract: Compare the applicability of three different statistical methods as multiple linear regression model,RBF neural network model and BP neural network model in the prediction of working age for silicosis Methods The databasewas established on the basis of stage⁃Ⅰsilicosis patients reported in Hebei province between 2006 and 2015, and the rank⁃sumtest was used for comparing the actual working age from the database and pridicted working ages by three models mentionedabove, and calculate their mean absolute error and mean relative error Results The results showed that a total number of stage⁃Ⅰsilicosis patients were 2 294 cases, the rank sum test showed that only the difference between actual value and predicted valuefrom RBF neural network model was statistically significant (P<0.05), the other two differences between actual value and predicted values from multiple linear regression model of BP neural network model were all no statistical significance (P>0. 05),and the BP neural network model had highest prediction accurac . Conclusion The results suggested that the selection of appropriate model should be based on data characteristics and analytical needs, and the BP neural network model seemed better than the multiple linear regression model in the prediction of working age for silicosis |
| Keywords: stage⁃I silicosis dust exposure duration multiple linear regression model RBF neural network model BP neural network model |
