期刊
ARTIFICIAL NEURAL NETWORKS AND MACHINE LEARNING - ICANN 2019: WORKSHOP AND SPECIAL SESSIONS
卷 11731, 期 -, 页码 420-425出版社
SPRINGER INTERNATIONAL PUBLISHING AG
DOI: 10.1007/978-3-030-30493-5_42
关键词
Bayesian; Neural networks; Hybrid Monte Carlo; Credit default modelling; Automatic Relevance Determination
This work develops a neural network based global model interpretation mechanism - the Bayesian Neural Network with Automatic Relevance Determination (BNN-ARD) for feature selection in credit default modelling. We compare the resulting selected important features to those obtained from the Random Forest (RF) and Gradient Tree Boosting (GTB). We show by re-training the models on the identified important features that the predictive quality of the features obtained from the BNN-ARD is similar to that of the GTB and outperforms those of RF in terms of the predictive performance of the retrained models.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据