BP神经网络与logistic回归的比较研究

The Research about the Comparison of Neural Network Model with logistic Regression

目的通过与logistic回归分析的比较,探讨BP神经网络在判别分析中的应用。方法设计合适的BP神经网络参数,采用LevenbergMarquardt优化算法来避免BP算法收敛速度慢的缺点,采用了“早停止”(earlystopping)策略避免过度拟合(overfitting),并把BP神经网络和logistic回归的结果作比较。结果BP神经网络在回代和前瞻性考核中都取得较好的结果,两者ROC曲线的比较也说明了这一点。结论BP神经网络方法值得在医学研究,特别是判别分析、生存分析领域进一步应用并推广。

Objective To explore the application of BP neural network on discriminant analysis through comparing with logistic regression model.Methods Levenberg-Marquardt algorithm is adopted which makes learning time short,convergence fast,early-stopping method is used for avoiding over-fitting.And compare the performance of a neural network model with that of logistic regression model.Results BP neuralnetwork gets good results in internal validation and external validation,the comparison of their ROC curves(relative operating characteristic curve) also give a good prove.Conclusion BP neural network is worthy to be popularized,especially in the fields of survival analysis and discriminant analysis.