Paper: arXiv 2611.18651

Authors: Kayode Lateef Ogunsusi, Mary Nwaife Mezue, Kazeem Busayo Ogunsusi, Adeniyi Paul Phillips

Abstract

Financial institutions increasingly require intelligent risk-management systems that combine predictive performance with data reliability, interpretability, and adaptable decision policies for emerging agentic commerce environments. This study proposes a Trusted Financial Data (TFD) Multi-Agent Framework that integrates predictive risk modeling, behavioral intelligence, data-trust assessment, explainable artificial intelligence, and uncertainty-aware decision orchestration for credit-default and payment-fraud management. The framework translates heterogeneous risk and trust signals into interpretable Approve, Review, and Decline decisions using fixed and adaptive decision policies. Experimental evaluation was conducted on the Home Credit Default Risk and IEEE-CIS Fraud Detection datasets against five machine-learning baselines: Logistic Regression, Random Forest, XGBoost, LightGBM, and Multi-Layer Perceptron. Adaptive TFD achieved recall values of 0.6234 and 0.9366, with ROC-AUC values of 0.7584 and 0.9474 on Home Credit and IEEE-CIS, respectively. Adaptive orchestration consistently improved minority-class risk detection relative to fixed TFD policies, although at the cost of increased false-positive interventions. SHAP-based global and local explanations further enhanced transparency into model predictions and decision behavior. Overall, the findings demonstrate the potential of TFD as a trust-aware multi-agent decision-orchestration layer that combines predictive, behavioral, uncertainty, and data-quality signals to support configurable, explainable, and adaptive financial risk management across heterogeneous operational environments.

Complexity vs Empirical Score

  • Math Complexity: 4.0/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper proposes a novel multi-agent framework for financial risk management, integrating various AI components. It demonstrates strong empirical evaluation on benchmark datasets, though the mathematical underpinnings are more applied than theoretical. The clarity and completeness are good, but reproducibility could be enhanced with explicit code/data availability.

Research Flowchart

  flowchart TD
    A["Research Goal: Develop Trusted Financial Data Framework for Agentic Commerce Risk Management"] --> B{"Key Methodology: Trusted Financial Data (TFD) Multi-Agent Framework"}
    B --> C["Data/Inputs: Home Credit Default Risk, IEEE-CIS Fraud Detection Datasets"]
    C --> D{"Computational Processes: Predictive Risk Modeling, Behavioral Intelligence, Data-Trust Assessment, XAI, Uncertainty-aware Decision Orchestration"}
    D --> E["Decision Policies: Fixed & Adaptive (Approve, Review, Decline)"]
    E --> F["Evaluation: Compare TFD (Fixed & Adaptive) vs. LR, RF, XGBoost, LightGBM, MLP"]
    F --> G["Key Findings/Outcomes: Adaptive TFD Improves Minority-Class Risk Detection (e.g., Recall: 0.6234 Home Credit, 0.9366 IEEE-CIS); XAI (SHAP) Enhances Transparency"]

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 a Trusted Financial Data (TFD) Multi‑Agent Framework that integrates predictive risk modeling, behavioral intelligence, data‑trust assessment, explainable AI, and adaptive decision orchestration for credit‑default and payment‑fraud management.

Source sentence
This study proposes a Trusted Financial Data (TFD) Multi-Agent Framework that integrates predictive risk modeling, behavioral intelligence, data-trust assessment, explainable artificial intelligence, and uncertainty-aware decision orchestration for credit-default and payment-fraud management.

Assumes

No explicit assumptions found in the text.

Data

  • Home Credit Default Risk benchmark public
    Source sentence
    The evaluation was conducted using the Home Credit Default Risk and IEEE-CIS Fraud Detection benchmark datasets. The Home Credit dataset contained 307,511 observations and 122 original variables, comprising demographic, financial, credit, and application-related attributes used to predict whether an applicant would experience payment difficulties.
  • IEEE‑CIS Fraud Detection benchmark public
    Source sentence
    The evaluation was conducted using the Home Credit Default Risk and IEEE-CIS Fraud Detection benchmark datasets. The IEEE-CIS dataset combines transaction and identity information using the TransactionID identifier.

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • Adaptive TFD achieved recall of 0.6234 on Home Credit and 0.9366 on IEEE‑CIS, with ROC‑AUC values of 0.7584 and 0.9474 respectively. performance
    Source sentence
    Adaptive TFD achieved recall values of 0.6234 and 0.9366, with ROC-AUC values of 0.7584 and 0.9474 on Home Credit and IEEE-CIS, respectively. On Home Credit, Adaptive TFD improved default Recall from 0.3837 to 0.6234 relative to Fixed TFD and achieved a ROC-AUC of 0.7584, comparable to XGBoost (0.7607) and LightGBM (0.7606).
  • Adaptive orchestration consistently increased minority‑class risk detection but incurred higher false‑positive interventions compared with fixed policies. performance
    Source sentence
    Adaptive orchestration consistently improved minority-class risk detection relative to fixed TFD policies, although at the cost of increased false-positive interventions. Across both datasets, adaptive orchestration consistently increased risk detection by applying more conservative interventions under greater data uncertainty, although this improvement was accompanied by reduced precision and increased
  • SHAP‑based global and local explanations enhanced transparency into model predictions and decision behavior. other
    Source sentence
    SHAP-based global and local explanations further enhanced transparency into model predictions and decision behavior.
  • The TFD framework is positioned as a trust‑aware decision‑orchestration layer rather than a replacement for high‑performing predictive models. theoretical
    Source sentence
    The findings therefore position TFD primarily as a trust-aware decision orchestration framework rather than a replacement for high-performing predictive models.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 14 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-17 · same card as JSON · A language model chose these items: check the source sentence before relying on one.