Paper: arXiv 2410.21928

Authors: Boris Wolfson, Erman Acar

Abstract

Current trends in Machine Learning prefer explainability even when it comes at the cost of performance. Therefore, explainable AI methods are particularly important in the field of Fraud Detection. This work investigates the applicability of Differentiable Inductive Logic Programming (DILP) as an explainable AI approach to Fraud Detection. Although the scalability of DILP is a well-known issue, we show that with some data curation such as cleaning and adjusting the tabular and numerical data to the expected format of background facts statements, it becomes much more applicable. While in processing it does not provide any significant advantage on rather more traditional methods such as Decision Trees, or more recent ones like Deep Symbolic Classification, it still gives comparable results. We showcase its limitations and points to improve, as well as potential use cases where it can be much more useful compared to traditional methods, such as recursive rule learning.

Complexity vs Empirical Score

  • Math Complexity: 4.5/10
  • Empirical Rigor: 5.5/10
  • Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper presents an innovative application of Differentiable Inductive Logic Programming (dILP) to fraud detection, bridging symbolic reasoning and machine learning for XAI. While it focuses heavily on data preprocessing adaptation rather than new mathematical derivations, its empirical setup on the PaySim dataset provides practical insights into performance trade-offs.

Research Flowchart

  flowchart TD
    A["Research Goal: Evaluate DILP for Explainable Fraud Detection"] --> B["Data Curation & Format Adaptation"]
    B --> C["Input: Tabular & Numerical Data as Background Facts"]
    C --> D["Computational Process: DILP Rule Learning"]
    D --> E["Performance & Scalability Evaluation vs. Decision Trees & DSC"]
    E --> F["Outcomes: Comparable Performance, Scalability Limits, & Recursive Rule Capacity"]

Cited References (35)

  1. A survey on neural-symbolic learning systems (2023)
  2. Rule Learning from Time-Dependent Data Applied to Fraud Detection (2021)
  3. Harnessing Deep Neural Networks with Logic Rules (2020) — link
  4. Anomaly detection with tensor networks (2020) — link
  5. Credit Card Transactions Fraud Detection Dataset (2020)
  6. Credit Card Transactions (2021)
  7. PaySim: A financial mobile money simulator for fraud detection (2016)
  8. ANNUAL FRAUD REPORT 2022 (2023)
and 27 more
  1. Empty streets, busy internet: A time-series analysis of cybercrime and fraud trends during COVID-19 (2021)
  2. Learning explanatory rules from noisy data (2018)
  3. Neuro-symbolic artificial intelligence: The state of the art (2022)
  4. Understanding Unfairness in Fraud Detection through Model and Data Bias Interactions (2022) — link
  5. Inductive logic programming (1991)
  6. Fraud detection in mobile payment systems using an XGBoost-based framework (2023)
  7. Explainable Fraud Detection with Deep Symbolic Classification
  8. Cart (1984)
  9. Learning accurate and interpretable decision rule sets from neural networks (2021)
  10. Binary cross entropy with deep learning technique for image classification (2020)
  11. Logic explained networks (2023)
  12. Symbolic data analysis as a tool for credit fraud detection (2022)
  13. A Comparative Study of Three Neural-Symbolic Approaches to Inductive Logic Programming (2022)
  14. Inductive logic programming via differentiable deep neural logic networks (2019) — link
  15. Abductive knowledge induction from raw data (2020) — link
  16. A financial fraud detection model based on LSTM deep learning technique (2020)
  17. Comparison of the predicted and observed secondary structure of T4 phage lysozyme (1975)
  18. Differentiable Rule Induction with Learned Relational Features (2022) — link
  19. Neural-based classification rule learning for sequential data (2022)
  20. Deep symbolic regression: Recovering mathematical expressions from data via risk-seeking policy gradients (2019) — link
  21. Representing biases for inductive logic programming (1994)
  22. DILP-Core github location (2020) — link
  23. Duck Db (2020) — link
  24. Decision trees do not generalize to new variations (2010)
  25. Performance or Explainability? A Law of Armed Conflict Perspective (2023)
  26. The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation (2020)
  27. The geometry of suspicious money laundering activities in financial networks (2022)

0 of 35 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.