| October 1, 2019
The goal of this project was to investigate the current techniques for explaining machine learning models, especially those suitable for fraud detection. The LIME and SHAP techniques have been studied when applied to random forests and multilayer perceptrons.
The original subject covered both in-store and online card payments, with particular attention to identifying the factors behind a rejected transaction. Such explanations can help bank staff understand a decision and investigate emerging fraud patterns. The students focused on feature attribution, comparing LIME’s local surrogate explanations [1] with SHAP’s feature contributions [2] on simulated transactions supplied by Lusis.
The study evaluated whether explanations recovered the rules used to generate fraudulent transactions, and compared them with decision paths in an interpretable decision tree. It also measured computation time, a practical constraint for real-time payment systems. The experiments compared SHAP’s generic KernelExplainer with TreeExplainer and DeepExplainer, illustrating the trade-off between execution time and agreement between explanations. Background reading supplied with the project covered model-agnostic interpretation, explanation quality and causal mechanisms [3-6].
References
[1] Marco Tulio Ribeiro, Sameer Singh and Carlos Guestrin. “Why Should I Trust You?”: Explaining the Predictions of Any Classifier. KDD, pp. 1135-1144, 2016. DOI.
[2] Scott M. Lundberg and Su-In Lee. A Unified Approach to Interpreting Model Predictions. Advances in Neural Information Processing Systems 30, pp. 4765-4774, 2017.
[3] Marco Tulio Ribeiro, Sameer Singh and Carlos Guestrin. Model-Agnostic Interpretability of Machine Learning. ICML Workshop on Human Interpretability in Machine Learning, 2016.
[4] Diogo V. Carvalho, Eduardo M. Pereira and Jaime S. Cardoso. Machine Learning Interpretability: A Survey on Methods and Metrics. Electronics 8(8), 832, 2019. DOI.
[5] Google. AI Explainability Whitepaper. 2019.
[6] Yoshua Bengio et al. A Meta-Transfer Objective for Learning to Disentangle Causal Mechanisms. arXiv:1901.10912, 2019.