Explainable Federated Learning for U.S. State-Level Financial Distress Modeling

TL;DR

This study applies cross-silo federated learning to predict debt-collection contact as a signal of consumer financial distress using NFCS survey data, treating each U.S. state as a data silo to preserve privacy. It introduces an 8-layer Highway Network with class weighting to handle severe data imbalance and integrates SHAP and Owen values to provide interpretable, global and state-specific explanations of predictors. Results show that the federated model achieves performance comparable to a centralized model while substantially reducing communication costs, and reveals meaningful state-level differences (e.g., Washington vs Hawaii) that can inform region-tailored interventions. The framework offers a scalable, regulation-compliant blueprint for socially responsible AI in consumer credit risk and financial inclusion, with broad applicability to privacy-sensitive decentralized data settings.

Abstract

We present the first application of federated learning (FL) to the U.S. National Financial Capability Study, introducing an interpretable framework for predicting consumer financial distress across all 50 states and the District of Columbia without centralizing sensitive data. Our cross-silo FL setup treats each state as a distinct data silo, simulating real-world governance in nationwide financial systems. Unlike prior work, our approach integrates two complementary explainable AI techniques to identify both global (nationwide) and local (state-specific) predictors of financial hardship, such as contact from debt collection agencies. We develop a machine learning model specifically suited for highly categorical, imbalanced survey data. This work delivers a scalable, regulation-compliant blueprint for early warning systems in finance, demonstrating how FL can power socially responsible AI applications in consumer credit risk and financial inclusion.

Figures (6)

• Figure 1: State-Level Evaluation of Federated Model Performance. Choropleth maps showing the performance of the global federated model across U.S. states using four evaluation metrics: precision, recall, F1-score, and AUC. Each state acts as an independent client in the cross-silo FL setup, revealing heterogeneity in model effectiveness and highlighting regional variation in financial distress prediction.
• Figure 2: Global performance metrics of the federated model over training rounds and final confusion matrix on the test set.
• Figure 3: State-level scatterplots of model performance metrics across all 51 clients (50 states + District of Columbia).
• Figure 4: Centralized Model Performance
• Figure 5: Global feature importance analysis using SHAP and Owen values.