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.
CITATION STYLE
Mbuvha, R., Boulkaibet, I., & Marwala, T. (2019). Bayesian Automatic Relevance Determination for Feature Selection in Credit Default Modelling. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11731 LNCS, pp. 420–425). Springer Verlag. https://doi.org/10.1007/978-3-030-30493-5_42
Mendeley helps you to discover research relevant for your work.