Attention-based dynamic multilayer graph neural networks for loan default prediction

Sahab Zandi, Kamesh Korangi, María Óskarsdóttir, Christophe Mues, Cristián Bravo

Abstract: Highlights•We model credit risk with deep learning leveraging dynamic multilayer networks.•We test our methodology in a behavioural credit scoring context.•Our custom attention mechanism weights networks according to their importance.•Our GAT-LSTM-ATT model outperforms other methods.