摘要
研究了Ⅱ型糖尿病并发症的预测。针对相关诊断指标众多,直接应用传统的神经网络等模型预测,会带来无法适应多种并发症、运算速度较慢及预测准确率偏低等问题,提出了基于Logistic回归和多层神经网络(MNN)的Ⅱ型糖尿病并发症预测模型。该模型首先应用关联性分析,提取与5种不同Ⅱ型糖尿病并发症相关的诊断指标,经Logistic回归模型等分析得到强相关因子,作为预测模型的输入,再运用Python,构建基于多层神经网络的预测模型。实验结果表明,全血糖化血红蛋白测定,尿胆原定性实验指标,尿素和尿红细胞与绝大部分Ⅱ型糖尿病并发症直接相关。Logistic回归结合多层神经网络预测准确率高于单一Logistic回归模型,预测准确率基本保持在85%的水平上,对某些并发症的预测准确率达到90%以上,可以达到为Ⅱ型糖尿病并发症预测提供科学参考的目的。
The prediction of type Ⅱ diabetes complications is studied. The diagnostic criteria of type Ⅱ diabetes complications are numerous, and direct application of traditional neural network model prediction will bring about the situation that it cannot adapt to many kinds of complications, slow operation speed and low prediction accuracy. A prediction model for type Ⅱ diabetes complications based on Logistic regression and multilayer neural network (MNN) is proposed. The model firstly uses correlation analysis to extract diagnostic indexes related to five different types of type Ⅱ diabetic complications, and the strong correlation factor is obtained through Logistic regression model analysis, and then a prediction model based on multilayer neural network is constructed by using Python language. The experiment results show that the measurement of total glycemic hemoglobin, urinary biliary original test indicators, urea and urinary red blood cells are related to most of the complications of type Ⅱ diabetes directly. The prediction accuracy of multi-layer neural network is higher than that of Logistic regression model. The prediction accuracies are maintained at 85% basically, and the prediction accuracy of certain complications can reach more than 90%, which provides a scientific reference for the complications of type Ⅱ diabetes.
作者
王洁
乔艺璇
彭岩
陈晓
Wang Jie;Qiao Yixuan;Peng Yan;Chen Xiao(School of Management, Capital Normal University, Beijing 100048;Department of Radiation Oncology, Air Force General Hospital,Beijing 100142)
出处
《高技术通讯》
EI
CAS
北大核心
2019年第5期455-461,共7页
Chinese High Technology Letters
基金
国家自然科学青年基金(61601310)
北京市教委社科(SM201910028017)
北京市教委科技创新服务能力建设(19530050142
19530050187)资助项目
