Model selection in neural networks

被引：228

作者：

Anders, U ^{[1
]}

Korn, O ^{[1
]}

机构：

[1] ZEW, Ctr European Econ Res, D-68034 Mannheim, Germany

来源：

NEURAL NETWORKS | 1999年 / 12卷 / 02期

关键词：

neural networks; statistical inference; model selection; identification; information criteria; cross validation; pruning; Monte Carlo simulation;

D O I：

10.1016/S0893-6080(98)00117-8

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

In this article, we examine how model selection in neural networks can be guided by statistical procedures such as hypothesis tests, information criteria and cross validation. The application of these methods in neural network models is discussed, paying attention especially to the identification problems encountered. We then propose five specification strategies based on different statistical procedures and compare them in a simulation study. As the results of the study are promising, it is suggested that a statistical analysis should become an integral part of neural network modeling. (C) 1999 Elsevier Science Ltd. All rights reserved.

引用

页码：309 / 323

页数：15

共 37 条

[1] NEW LOOK AT STATISTICAL-MODEL IDENTIFICATION
AKAIKE, H
[J]. IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 1974, AC19 (06) : 716 - 723
[2] Akaike H., 1973, 2 INT S INFORM THEOR, P267, DOI [DOI 10.1007/978-1-4612-1694-0_15, 10.1007/978-1-4612-1694-0_15]
[3] ANDERS U, 1998, IN PRESS J FORECASTI
[4] ANDERS U, 1997, PROGR CONNECTIONIST, P1
[5] [Anonymous], 1997, INTELLIGENT SYSTEMS
[6] [Anonymous], 1993, J. Time Ser. Anal., DOI DOI 10.1111/J.1467-9892.1993.TB00139.X
[7] Bishop C. M., 1995, NEURAL NETWORKS PATT
[8] BURGESS AN, 1995, IEE P 4 INT C ART NE, V409, P312
[9] Davidson R., 1993, ESTIMATION INFERENCE, DOI DOI 10.1017/S0266466600009452
[10] HYPOTHESIS TESTING WHEN A NUISANCE PARAMETER IS PRESENT ONLY UNDER ALTERNATIVE
DAVIES, RB
[J]. BIOMETRIKA, 1977, 64 (02) : 247 - 254

← 1 2 3 4 →