Interpretability-by-Design with Accurate Locally Additive Models and Conditional Feature Effects
This work addresses the trade-off between interpretability and accuracy in machine learning models for practitioners needing auditable predictions, though it is incremental as it builds on existing GAM and GA²M frameworks.
The paper tackled the problem of balancing interpretability and accuracy in generalized additive models (GAMs) by proposing Conditionally Additive Local Models (CALMs), which allow multiple univariate shape functions per feature in different input regions to capture interactions, resulting in accuracy comparable to GA²Ms while maintaining interpretability.
Generalized additive models (GAMs) offer interpretability through independent univariate feature effects but underfit when interactions are present in data. GA$^2$Ms add selected pairwise interactions which improves accuracy, but sacrifices interpretability and limits model auditing. We propose \emph{Conditionally Additive Local Models} (CALMs), a new model class, that balances the interpretability of GAMs with the accuracy of GA$^2$Ms. CALMs allow multiple univariate shape functions per feature, each active in different regions of the input space. These regions are defined independently for each feature as simple logical conditions (thresholds) on the features it interacts with. As a result, effects remain locally additive while varying across subregions to capture interactions. We further propose a principled distillation-based training pipeline that identifies homogeneous regions with limited interactions and fits interpretable shape functions via region-aware backfitting. Experiments on diverse classification and regression tasks show that CALMs consistently outperform GAMs and achieve accuracy comparable with GA$^2$Ms. Overall, CALMs offer a compelling trade-off between predictive accuracy and interpretability.