基于学习向量量化网络建立糖尿病并发症预测模型

Establishing a Model for Predicting Diabetes Complications Based on the LVQ Neural Work

目的建立基于人工神经网络的糖尿病并发症预测模型。方法选择学习向量量化（Learn Vector Quatization，LVQ）人工神经网络，均衡分配临床诊断数据集进行训练和测试，建立LVQ网络框架并调整其隐含层神经元数目、线性层权重比例、学习速率和设置最大循环次数。在Matlab6．5的基础上编写程序，将神经网络的预测结果与实际结果进行比较。结果对于5种糖尿病慢性并发症的预测符合率在64．71％。82．35％。结论神经网络理论对数据的要求较低，容错能力强，且可以将人们对客观世界的感受和判定与定理建模技术结合起来，提高了判别的准确性，适合医学领域的数据挖掘应用，这将有利于人工神经网络技术发展也有利于医学拓展自身的模式空间。

Objective To build up a predictive model of the diabetes complications according to the LVQ nerve network. Methods In tiffs paper, we had a detailed discussion about how to allot data to carry on train and test in balance, build up a network, adjnst the number of S1, and how to design the weights for neural network and so on. Results The model match a rate of 64. 71% - 82.35%. Conclusion Neural network is worthy to be popularized in medicine.