Paper: arXiv 2511.08588
Authors: Lorenzo Carta, Fernando Spadea, Oshani Seneviratne
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.
Complexity vs Empirical Score
- Math Complexity: 6.0/10
- Empirical Rigor: 7.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced distributed machine learning (federated learning) and interpretable AI techniques (SHAP, Owen values) but is grounded in a real-world dataset (NFCS) with detailed methodology, performance metrics, and a provided GitHub repository.
Research Flowchart
flowchart TD A["Research Goal: Predict U.S. State-Level<br>Financial Distress via Federated Learning"] --> B["Data: National Financial Capability Study<br>50 States + DC"] B --> C["Method: Cross-Silo Federated Learning<br>Each State = Silo"] C --> D["Computations: Privacy-Preserving<br>Model Training & Weight Aggregation"] D --> E["Key Outcomes:<br>1. Global & Local Explanations<br>2. State-Specific Risk Patterns<br>3. Scalable FL Framework"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes the first federated learning application to predict consumer financial distress across all U.S. states with an interpretable framework that does not centralize sensitive data.
Source sentence
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.
Assumes 5
- Each U.S. state can be treated as an independent data silo for cross‑silo federated learning. data
Source sentence
Our cross-silo FL setup treats each state as a distinct data silo, simulating real-world governance in nationwide financial systems.
- Class weighting can effectively mitigate the severe class imbalance (only 18% minority class). model
Source sentence
using class weighting to address the significant imbalance (only 18% of respondents had been contacted by a DCA)
- A minority class weight factor of 1.1835 yields the best precision‑recall trade‑off. statistical
Source sentence
increasing the standard minority class weight by a factor of 1.1835 yielded the best balance between precision and recall
- The 2021 NFCS survey responses are representative of state‑level financial distress patterns. data
Source sentence
more than 25,000 U.S. respondents, with approximately 500 respondents per state
Show 1 more
- Communication cost reductions (GB transferred) are a relevant metric for federated learning efficiency. execution
Source sentence
We also optimized model update communication, reducing costs from 12.954 GB to 4.935 GB over 200 training rounds.
Data
- 2021 National Financial Capability Study (NFCS) alternative public
Source sentence
for our analysis. This comprehensive dataset is derived from a 126‑question multiple choice survey that captures various statistics such as income, education, and financial education from more than 25,000 U.S. respondents, with approximately 500 respondents per state, plus the District of Columbia.
Costs
Not applicable
Finds 5
- Federated learning achieves predictive performance on‑par with a centralized model and outperforms locally trained state models. predictive-accuracy
Source sentence
showed through comparison that our results were on-par with the centralized model, while also outperforming localized ones.
- Communication overhead was cut by more than 60%, from 12.954 GB to 4.935 GB across 200 rounds. performance
Source sentence
We also optimized model update communication, reducing costs from 12.954 GB to 4.935 GB over 200 training rounds.
- Critical predictors of financial distress include Gender/Age and Annual Income. statistical
Source sentence
Our analysis identified critical predictors of financial distress, including Gender/Age and Annual Income.
- Combining SHAP and Owen values enables interpretation of both global (national) and local (state‑level) model behavior. theoretical
Source sentence
Our contribution is distinct in combining SHAP with Owen values, enabling interpretation of both global (national) and local (state-level) model behavior.
- A minority class weight factor of 1.1835 provides the best balance between precision and recall. statistical
Source sentence
increasing the standard minority class weight by a factor of 1.1835 yielded the best balance between precision and recall
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 37 related papers). Coverage is partial, so this is neither replication nor refutation.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (21)
- Highway and residual networks learn unrolled iterative estimation (2016) — link
- The FINRA Foundation’s National Financial Capability Study: Unpacking 12 years of data on US financial capability (2024)
- Communication-efficient learning of deep networks from decentralized data (2017)
- Fed-RD: Privacy-preserving federated learning for financial crime detection (2024)
- Federated machine learning: Concept and applications (2019)
- Federated Learning for Credit Risk Assessment. (2023)
- A novel federated learning approach with knowledge transfer for credit scoring (2024)
- Financial Data Analysis with Robust Federated Logistic Regression — → 2504.20250
and 13 more
- Financial distress prediction in an international context: A review and empirical analysis of Altman’s Z-score model (2017)
- Financial distress prediction models: a review of their usefulness 1 (2019)
- Using Machine Learning and Counterfactual Explanations for Financial Distress Prediction (2024)
- The many Shapley values for model explanation (2020)
- Approximation of group explainers with coalition structure using Monte Carlo sampling on the product space of coalitions and features (2023) — link
- Secure and transparent banking: explainable AI-driven federated learning model for financial fraud detection (2025)
- Robust integration of blockchain and explainable federated learning for automated credit scoring (2024)
- Adam: A method for stochastic optimization (2014) — link
- Access to Credit and Financial Health: Evaluating the Impact of Debt Collection (2017) — link
- The Number of Americans with Debt in Collections Fell during the Pandemic to 64 Million (2022) — link
- Flower: A friendly federated learning research framework (2020) — link
- Language modeling with highway LSTM (2017)
- Empower sequence labeling with task-aware neural language model (2018)
1 of 21 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.