| October 1, 2021
Project Overview
The project focuses on improving the explainability of fraud detection models used in banking, an industry where financial losses due to fraudulent transactions amount to an estimated $25 billion annually. As banks increasingly rely on machine learning (ML) algorithms for fraud detection, ensuring these models are interpretable is crucial for transparency, especially when justifying false positive decisions to customers.
Existing post-hoc explainability methods, such as LIME [1] and ANCHOR [2], struggle with fraud detection due to the extreme imbalance of fraud datasets, where fraudulent transactions are a small minority. This imbalance makes traditional sampling-based approaches unreliable, leading to weak and inconsistent explanations.
The project aims to explore alternative rule-based explainability methods. One approach, proposed in [3], generates interpretable decision sets that balance predictive accuracy with interpretability. Another method, presented in [4], constructs ordered rule lists, providing structured and sequential decision-making explanations.
To evaluate these methods, the project involves implementing the models described in [3] and [4] and comparing their effectiveness using established evaluation metrics. A key improvement involves modifying the method from [3] to enhance the diversity of categorical feature values, ensuring broader and more meaningful rule sets.
By advancing explainability in fraud detection, the project seeks to help banks implement transparent AI-driven decision-making while maintaining high detection accuracy. This will ultimately improve customer trust and regulatory compliance in the banking sector.
References
[1] Marco Tulio Ribeiro and Sameer Singh and Carlos Guestrin, . “Why Should I Trust You?”: Explaining the Predictions of Any Classifier. . In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, CA, USA, August 13-17, 2016 (pp. 1135–1144).2016.
[2] Marco Tulio Ribeiro, , Sameer Singh, and Carlos Guestrin. Anchors: High-Precision Model-Agnostic Explanations. . In AAAI Conference on Artificial Intelligence (AAAI). 2018.
[3] Lakkaraju, Himabindu, Stephen H., Bach, and Jure, Leskovec. Interpretable Decision Sets: A Joint Framework for Description and Prediction. . In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 1675–1684). Association for Computing Machinery, 2016.
[4] Aoga, Pierre. Finding Probabilistic Rule Lists using the Minimum Description Length Principle. . In Discovery Science (pp. 66–82). Springer International Publishing, 2018.