Base neural network models for credit scoring of individuals

▤ Summary

This article discusses neural network base models for individual credit scoring. It compares model architectures such as RNN-based encoders, Transformer variants, and CLS-token approaches for supervised prediction. The text also reviews training choices like hard negative mining, loss functions, normalization, pruning, and optimizer settings. The main outcome is that the RNN feature extractor performs best on the reported Gini metric. The piece reads like a technical research summary rather than a company announcement.

▦ Linked Companies