目的 采用数据挖掘方法, 考察2型糖尿病的危险因素, 确定最优风险预测模型, 为建立手机APP软件提供算法, 为糖尿病I级预防提供风险预测支持.方法 收集某医院2016年1月—2017年7月的糖尿病患者全数据集, 共5 571例, 通过与同期体检健康对照组5 571例进行对比研究, 分别建立Logistic回归模型和多层感知器神经网络模型, 比较优劣, 确定最终预测模型.结果 结果显示Logistic回归和多层感知器神经网络模型对训练样本的预测准确率分别为89.7％、80.4％, 对测试样本的预测准确率分别为89.8％、79.8％.结论 Logistic回归模型对2型糖尿病风险预测效能较高, 预测结果也更容易结合临床实际, 用于风险控制手机APP软件后台编程.%Objective To investigate the risk factors of type 2 diabetes by using data mining methods, to determine the optimal risk prediction model, to provide algorithms for establishing mobile APP software, and to provide risk prediction support for diabetes level I prevention. Methods A total of 5 571 patients with diabetes mellitus from January 2016 to July 2017 in the hospital were enrolled. A logistic regression model and a multi-layer perceptron neural network model were established by comparing with 5 571 healthy people in the same period, comparing the pros and cons, determine the final prediction model. Results The results showed that the prediction accuracy of the logistic regression and multi-layer perceptron neural network model for training samples were 89.7％ and 80.4％, respectively, and the prediction accuracy for the test samples was 89.8％ and 79.8％, respectively. Conclusion Logistic regression model has higher predictive effect on risk of type 2 diabetes, and the prediction results are more easily combined with clinical practice. It is used for background programming of risk control mobile APP software.
展开▼
