Neural-Network-Based Multi-Agent Learning for Credit Scoring

In this study, a neural-network-based multi-agent learning approach is proposed for credit risk evaluation. First of all, a specified credit dataset is divided into several non-overlapping subsets, i.e., training, validation and testing sets. Then some feed-forward neural networks (FNNs) with different topological architectures are used as intelligent-agents to model the dataset and the corresponding classification scores can be obtained. Finally, all classification scores obtained are fused into an aggregated score as the final classification results using an adaptive linear neural network (ALNN). For illustration, a practical credit dataset is used to verify the effectiveness of the proposed neural-network-based multi-agent learning paradigm.