Interpretable vs Learned Encoders for High-Cardinality Fraud Detection

2026-07-01

AI Summary

This study addresses the challenge of balancing model performance and interpretability when handling high-cardinality categorical features in fraud detection. The authors systematically evaluate seven encoding strategies—including entity embeddings, target encoding, and hierarchical grouping—on the IEEE-CIS dataset using LightGBM as the primary model, with comparative analyses against CatBoost and TabNet. Empirical results are obtained through stratified 5-fold cross-validation with three repetitions. Entity embeddings achieve the highest AUC-ROC (0.9612), matching CatBoost’s performance and significantly outperforming hierarchical encoding; CatBoost leads in AUC-PR (0.822); TabNet underperforms under data sparsity. The findings highlight the benefits of joint multi-column representations and provide empirical guidance for selecting encoding methods for high-cardinality features in fraud detection tasks.

Abstract

A total of seven categorical encoding methods were tested on the IEEE-CIS fraud benchmark dataset (590,540 records, 3.5% positives, 8 high-cardinality columns). The encoders were evaluated using a stratified 5-fold cross-validation (CV) with three repetitions. Five of the encoders had identical frozen LightGBM learners in the downstream phase, allowing for controlled comparisons of their performance to each other. CatBoost and TabNet were included as comparisons across paradigms using different learners. The entity embeddings produced the highest AUC-ROC (0.9612), with a statistically significant tie with that of CatBoost (0.9602) and statistically superior to tier group encoding (0.9548), whereas target encoding was only 0.0023 worse than tier group encoding and the auditor-friendly tier boundaries were maintained. Off-the-shelf TabNet did not outperform tree-based pipelines and collapsed under data scarcity. On AUC-PR, CatBoost leads (0.822 vs. 0.793); no encoder dominated both metrics. Per-column analysis confirmed the embedding advantage arises from joint multi-column representation.

Problem

• high-cardinality
• fraud detection
• categorical encoding
• interpretability
• AUC-ROC

Innovation

• entity embeddings
• high-cardinality encoding
• interpretable encoding
• fraud detection
• controlled encoder comparison