Interpretability-by-Design with Accurate Locally Additive Models and Conditional Feature Effects

2026-02-18

AI Summary

This work proposes Conditional Additive Local Models (CALMs) to address the limitations of Generalized Additive Models (GAMs), which underfit in the presence of feature interactions due to their restriction to univariate effects, and GA²Ms, which improve accuracy by incorporating explicit interaction terms at the cost of interpretability. CALMs unify local additivity with global interactions by activating multiple local univariate shape functions conditioned on simple logical rules over different input regions. The method employs a distillation-guided training procedure to identify low-interaction homogeneous regions and fits interpretable shape functions using a region-aware backfitting algorithm. Experiments demonstrate that CALMs significantly outperform GAMs across diverse classification and regression tasks, achieving prediction accuracy comparable to GA²Ms while preserving strong interpretability.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Graphical Models
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

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.

Problem

• interpretability
• feature interactions
• generalized additive models
• model accuracy
• local additivity

Innovation

• Conditionally Additive Local Models
• Interpretability-by-Design
• Local Additivity
• Feature Interactions
• Distillation-based Training