AI Insight
Researchers developed Conditionally Additive Local Models (CALMs), a new machine learning approach that balances model interpretability with predictive accuracy. Unlike traditional additive models that assume feature effects are constant everywhere, CALMs allow feature effects to vary across different regions of the data space defined by simple logical conditions, enabling the model to capture interactions while maintaining interpretability through univariate shape functions. Experimental results demonstrate that CALMs consistently outperform standard Generalized Additive Models (GAMs) and achieve accuracy comparable to more complex GA2Ms while preserving the ability to audit individual feature effects.
Why it matters
This approach addresses a fundamental challenge in machine learning: the trade-off between model accuracy and interpretability. CALMs could be particularly valuable in regulated domains like healthcare, finance, and criminal justice where both accurate predictions and transparent decision-making explanations are required.
Understand the Science
⚠️ Preprint – Noch nicht peer-reviewed
Dieser Artikel wurde noch nicht von unabhängigen Experten begutachtet. Die Ergebnisse sind vorläufig und sollten mit Vorsicht interpretiert werden.
Abstract: Generalized additive models (GAMs) offer high 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. We propose Conditionally Additive Local Models (CALMs), a new model class that narrows this trade-off: it stays close to GAMs in interpretability and close to GA$^2$Ms in accuracy. 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 broadly comparable to GA$^2$Ms, while preserving the univariate auditability that GA$^2$Ms forfeit.
Source: Interpretability-by-Design with Accurate Locally Additive Models and Conditional Feature Effects