Challenging the Performance-Interpretability Trade-off: An Evaluation of Interpretable Machine Learning Models

TL;DR

The misconception that only black-box models can achieve high accuracy for tabular data is dispels by demonstrating that there is no strict trade-off between predictive performance and model interpretability for tabular data.

Abstract

Machine learning is permeating every conceivable domain to promote data-driven decision support. The focus is often on advanced black-box models due to their assumed performance advantages, whereas interpretable models are often associated with inferior predictive qualities. More recently, however, a new generation of generalized additive models (GAMs) has been proposed that offer promising properties for capturing complex, non-linear patterns while remaining fully interpretable. To uncover the merits and limitations of these models, this study examines the predictive performance of seven different GAMs in comparison to seven commonly used machine learning models based on a collection of twenty tabular benchmark datasets. To ensure a fair and robust model comparison, an extensive hyperparameter search combined with cross-validation was performed, resulting in 68,500 model runs. In addition, this study qualitatively examines the visual output of the models to assess their level of interpretability. Based on these results, the paper dispels the misconception that only black-box models can achieve high accuracy by demonstrating that there is no strict trade-off between predictive performance and model interpretability for tabular data. Furthermore, the paper discusses the importance of GAMs as powerful interpretable models for the field of information systems and derives implications for future work from a socio-technical perspective.

Figures (12)

• Figure 1: Comparison of shape functions of a linear model and a .
• Figure 2: Illustration of our evaluation strategy using 5-fold (stratified) cross-validation for error estimation and inner train-validation splitting for hyperparameter tuning with final retraining in each iteration.
• Figure 3: Global and local interpretability of an model. (1) The summary plots visualize the overall feature importance and the global relationships between the input features and the prediction target. The red dots illustrate the possibility of reading local values of individual feature effects for an exemplary sample. (2) The corresponding calculation example shows the additive contributions of each feature to the model output to retrieve the final prediction for the selected sample.
• Figure 4: Comparison of shape plots learned by five different and a linear model.
• Figure 5: Summary of the performance-interpretability evaluation. The x-axis represents the interpretability score from the interpretability assessment. The y-axis represents the performance score, computed by inverting the average rank of each model.