Bias reduction in skewed binary classification with Bayesian neural networks

被引：15

作者：

Lisboa, PJG ^{[1
]}

Vellido, A ^{[1
]}

Wong, H ^{[1
]}

机构：

[1] Liverpool John Moores Univ, Sch Comp & Math Sci, Liverpool L3 3AF, Merseyside, England

来源：

NEURAL NETWORKS | 2000年 / 13卷 / 4-5期

关键词：

calibration; marginalisation; Bayesian neural networks; bias;

D O I：

10.1016/S0893-6080(00)00022-8

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 [模式识别与智能系统]; 0812 [计算机科学与技术]; 0835 [软件工程]; 1405 [智能科学与技术];

摘要：

The Bayesian evidence framework has become a standard of good practice for neural network estimation of class conditional probabilities. In this approach the conditional probability is marginalised over the distribution of network weights, which is usually approximated by an analytical expression that moderates the network output towards the midrange. In this paper, it is shown that the network calibration is considerably improved by marginalising to the prior distribution. Moreover, marginalisation to the midrange can seriously bias the estimates of the conditional probabilities calculated from the evidence framework. This is especially the case in the modelling of censored data. (C) 2000 Elsevier Science Ltd. All rights reserved.

引用

页码：407 / 410

页数：4

共 8 条

[1]

Biganzoli E, 1998, STAT MED, V17, P1169, DOI 10.1002/(SICI)1097-0258(19980530)17:10<1169::AID-SIM796>3.3.CO

[2]

2-4

[3]

Regularization in skewed binary classification [J].