基于Logistic回归变量筛选的BP神经网络及应用

The Analysis and Application of BP Neutral Networks Based on Logistic Regression Variable Selection

BP神经网络是一种使用非线性可导函数作为传递函数的前馈神经网络,具有较高的精确度,但过多的预测变量会影响BP神经网络的准确性。采用Logistic回归变量筛选方法能在一定程度上提高分类准确性,提高模型效率。对2013年沪深两市A股分类评级进行了研究,证明基于Logistic回归变量筛选的神经网络提高了两极类别分类的准确性。

BP neutral networks is a sort of feedforward nertral networks with high accuracy ,which uses a supervised learn‐ing and nolinear differentiable function as the transfer function .However in the condition of excessive predictive variables , the neutral networks’accuracy would decrease to a considerable low level .With the introduction of logistic regression vari‐able selection ,the accurary and efficiency of the classification of BP neutral networks would increase to some extent . Through the research of the classification of Shanghai and Shenzhen stock market based on the statistics in 2013 ,the con‐clusion is reached ,that BP neutral networks based on logistic regression variable selection raises the accuracy of the classi‐fication in two polar catagories .