模型参数对疾病发病率人工神经网络模型精度的影响

THE INFLUENCE OF PARAMETERS ON PREDICTION ACCURACY OF ANN MODEL FOR THE INCIDENCE OF A DISEASE

本文通过统计分析 ,选取影响银川地区疾病发病率的主要气象因素 ,将其作为输入变量经多层前馈型神经网络的BP算法进行学习训练 ,建立了疾病发病率的人工神经网络 (ANN)预报模型。分析了结构参数对模型效果的影响情况 ,发现学习率和动量因子对达到训练目的无大的影响 ,而训练精度、输入层节点数和隐含层节点数是模型的关键。但只要输入层节点数达到一定数量 ,改变输入层节点数并不影响模型质量 。

This paper selects those elements which have significant effect on diseases by statistical analysis in Yinchuan area and sets up ANN model for the incidence of the diseases by putting those elements into many-level BP arithmetic of Feed-forward Backprop neural network as input varieties to study for training.And the outcome of it is compared with that of statistical model.The influence of model parameters on the fitting and prediction accuracy of model was studied.It can be concluded that the learning speed and momentum elements are little useful to the training goal,but the number of input and hidden notes are important.While the input number is up to a definite number,it won't influence the quality of model to change the input nodes.But the hidden nodes is more important to the model.