基于变量聚类的BP神经网络术后生存期预测模型

针对结直肠癌患者术后生存期预测，基于模糊C均值（FCM）聚类算法，提出一种结合场景认知和隶属度排序的变量聚类方法，对结直肠癌患者样本进行降维，并筛选出6个特征变量.结合BP神经网络，建立一个结直肠癌患者术后生存期预测模型.为了验证该模型的有效性，利用主成分分析（PCA）对样本进行降维，并训练BP神经网络，对比FCM模型及PCA模型的预测准确率.结果显示，基于FCM变量聚类的BP神经网络模型预测准确率更高，所提出的变量聚类方法能够有效筛选出对于生存期有相关性和解释性的变量，从而提高BP神经网络模型的预测准确率.
Abstract: A variable cluster method combining scenario cognition and membership degree ranking was proposed based on fuzzy C-means (FCM) cluster algorithm aiming at the postoperative survival prediction of colorectal cancer patients. Dimension reduction on samples of colorectal cancer patients were conducted; six characteristic variables were selected. Next, a postoperative survival prediction model was constructed for colorectal cancer patients with BP neural network. To verify the validity of this model, principal component analysis (PCA) was used to reduce the dimensions of the sample to train a BP neural network, and the comparison of accuracy rates between models based on FCM and PCA was conducted. Results verifly that the BP neural network model based on FCM variable cluster has more accurate prediction rate. The proposed variable cluster method can effectively screen out variables that have high pertinence and good interpretability of survival time, thus improves the forecast accuracy of BP neural network model.