Improving Credit Card Fraud Detection with an Optimized Explainable Boosting Machine

Addressing class imbalance is a central challenge in credit card fraud detection, as it directly impacts predictive reliability in real-world financial systems. To overcome this, the study proposes an enhanced workflow based on the Explainable Boosting Machine (EBM)-a transparent, state-of-the-art implementation of the GA2M algorithm-optimized through systematic hyperparameter tuning, feature selection, and preprocessing refinement. Rather than relying on conventional sampling techniques that may introduce bias or cause information loss, the optimized EBM achieves an effective balance between accuracy and interpretability, enabling precise detection of fraudulent transactions while providing actionable insights into feature importance and interaction effects. Furthermore, the Taguchi method is employed to optimize both the sequence of data scalers and model hyperparameters, ensuring robust, reproducible, and systematically validated performance improvements. Experimental evaluation on benchmark credit card data yields an ROC-AUC of 0.983, surpassing prior EBM baselines (0.975) and outperforming Logistic Regression, Random Forest, XGBoost, and Decision Tree models. These results highlight the potential of interpretable machine learning and data-driven optimization for advancing trustworthy fraud analytics in financial systems.

💡 Research Summary

This paper addresses the critical challenge of class imbalance in credit card fraud detection by proposing an enhanced and optimized workflow centered on the Explainable Boosting Machine (EBM). The authors identify that while machine learning is crucial for fraud analytics, many high-performing models are “black boxes,” and traditional sampling techniques to handle imbalance can introduce bias or information loss. Their solution strategically avoids resampling and instead focuses on optimizing a transparent, inherently interpretable model.

The core methodology employs the Explainable Boosting Machine, a state-of-the-art implementation of the Generalized Additive Models with Interactions (GA²M). EBM provides high accuracy comparable to ensemble methods like Random Forest and XGBoost while offering full interpretability through visualizations of individual feature contributions and pairwise interaction effects. To maximize EBM’s performance on the highly imbalanced European card fraud dataset (with a 1:577 fraud-to-legitimate ratio), the authors integrate the Taguchi method—a design of experiments approach—into their pipeline. This method is used to systematically and efficiently optimize two key aspects simultaneously: the sequential order of applying different data scalers (e.g., MinMaxScaler, StandardScaler) during preprocessing, and the hyperparameters of the EBM model itself. This approach ensures robust and reproducible performance gains with minimal experimental runs.

A notable aspect of the exploratory data analysis is the use of Chatterjee’s correlation coefficient (ξ) alongside traditional Pearson and Spearman metrics. This allows the detection of non-monotonic, functional relationships between features that standard linear correlation measures might miss, providing deeper insights into the data structure.

The experimental results demonstrate the effectiveness of the proposed framework. The optimized EBM model achieved an ROC-AUC score of 0.983 on the benchmark dataset. This performance not only surpasses the prior EBM baseline (0.975) but also outperforms several commonly used models, including Logistic Regression, Random Forest, XGBoost, and Decision Trees. Beyond the superior predictive accuracy, the model maintains its core advantage of interpretability, allowing analysts to understand which features (like V17, as indicated) are most influential in flagging a transaction as fraudulent.

In conclusion, this research presents a compelling case for combining interpretable machine learning models with systematic optimization techniques in high-stakes domains like financial fraud detection. By forgoing conventional sampling and leveraging the Taguchi method for hyperparameter and preprocessing optimization, the authors developed an EBM-based solution that successfully balances high predictive power with the transparency necessary for real-world trust and actionable insights. The work highlights the potential of “glass-box” models and data-driven optimization to advance reliable and accountable analytics in the financial sector.