三种分类预测模型在医学中的应用研究

A Study on Application of Three Prediction Models in Medicine

基于一个肾衰竭患者数据,应用两种神经网络(BP神经网络和贝叶斯正则化BP神经网络)与常用的二分类Logistic回归对肾衰竭患者是否死亡进行预测,并比较三种模型的预测效果.三个模型的判对率都达到89%以上.其中,以贝叶斯正则化BP神经网络的判对率和ROC曲线下面积(AUC)最大,即预测效果最好;BP神经网络和Logistic回归预测效果差别不大.

Based on a real dataset, we compare two kinds of neural network（ BP Neural Network and Bayesian Regularized BP Neural Network） and Logistic Regression in medical statistics. By using SPSS 21. 0 and Matlab, after variable screening,train the three models and compare their prediction accuracy. Besides, draw their ROC curves and compare their areas under the curve（ AUC） among the three models. All of the three models have reached the prediction accuracy over 89%. Bayesian Regularized BP Neural Network has the best results with the highest prediction accuracy and the largest AUC. Unlike the researches before,in our study,BP Neural Network did not have a better performance than Logistic Regression. The small sample size may result in the BP Neural Network without a good training. However,it may also highlight an advantage of Bayesian Regularized BP Neural Network,which still gets a good output under the situation of a small sample.