Complexity-Budgeted, Interaction-Aware Interpretable Model for Tabular Data - Abstract & Introduction

Sep 21, 2026

Inherently interpretable classifiers for tabular data often depend on sparse features, rules, or patterns that users can inspect directly. A key weakness is marginal feature screening, which can discard variables whose predictive power appears only through joint configurations. Interaction Aware Interpretable Machine Learning (IAIML) mitigates this via adaptive discretization, pairwise interaction scoring on a finite grid, and a partitioned explanation budget. On a 40-dataset panel, IAIML matches tuned ensembles in AUC while using far fewer explanation components, and it degrades mainly when higher-order interactions dominate.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages31

Reading Time1 h 18 min

AccessFree to download

Abstract

1 Introduction

Interpretability requirements in high-stakes domains

Vulnerability of marginal feature screening to interaction-only signal

Limitations in existing interpretable model families

arXiv :2607 .07060v 1 [ cs .LG] 8 Jul 2026 Complexity-Budgeted, Interaction-Aware Interpretable Model for Tabular Data Srikumar Krishnamoorthy Information Systems Area, Indian Institute of Management Ahmedabad, India Abstract Inherently interpretable classifiers for tabular data typically rely on sparse features, rules, or patterns that users can inspect directly. The marginal feature-screening step common to these methods can discard variables whose predictive value emerges only through joint configurations with other variables. We present Interaction Aware Interpretable Machine Learning (IAIML), a framework that addresses this limitation through three coordinated mechanisms: adaptive per-feature discretization, finite-grid pairwise interaction scoring, and a partitioned explanation budget. Detected interactions are routed through one of two strategies: relaxing the screening filter so that interaction-supported variables enter the pattern search, or constructing explicit pair terms for a sparse downstream classifier. On a 40-dataset panel comprising 24 real-world tabular benchmarks and 16 synthetic interaction stress tests, evaluated under nested cross-validation, IAIML achieves mean AUC within 1.4 points of tuned gradient-boosted ensembles while requiring roughly 14–28 times fewer fitted explanation components. On datasets with strong pairwise interaction structure and low marginal signal, IAIML outperforms all baselines. Among compact interpretable methods, IAIML is comparable to RuleFit in AUC and component count and is less expensive to tune. EBM obtains a small but significant AUC advantage across the full panel, with a substantially larger lookup-table footprint. Performance degrades on datasets requiring higher-order interactions beyond the pairwise scope. Component-isolated ablations confirm that adaptive discretization and interaction-aware admission each contribute incrementally. These results support IAIML as a compact, interaction-aware framework appropriate for settings where bounded explanation size and controlled treatment of feature interactions are design requirements. Keywords: Interpretable machine learning, Feature interaction detection, Pattern mining, Classification, Complexity budgeting 1 Introduction Machine learning is increasingly used to support important decisions in healthcare, finance, criminal justice, and public policy. In these settings, stakeholders require that automated decisions be explainable, auditable, and contestable (Rudin, 2019; Guidotti et al. , 2019; De Bock et al. , 2024) . This requirement has motivated a growing body of work on inherently interpretable models, which provide transparency by construction, as opposed to post-hoc explanation methods that approximate opaque models after the fact (Ribeiro et al. , 2016; Lundberg and Lee, 2017) . Interpretable model families include decision trees (Breiman et al. , 1984), rule lists (Letham et al. , 2015; Angelino et al. , 2018), scoring systems (Ustunand Rudin, 2016), generalized additive models (Lou et al. , 2012 , 2013), and pattern-based classifiers (Krishnamoorthy, 2024) . A common thread across these families is sparsity: the model’s prediction is traceable to a small number of features, rules, or patterns that a human reviewer can inspect directly. Achieving sparsity, however, typically requires an early screening step that ranks candidate features by their individual association with the target variable. This marginal screening is a natural choice for auditability and computational control, but it introduces a systematic vulnerability. Two variables may each be individually uninformative while their joint configuration is highly predictive of the outcome. Consider a clinical setting in which a patient’s risk depends on the combination of two biomarkers, neither of which is predictive on its own. A screening rule that ranks features by marginal association will discard both biomarkers before their joint sign

FAQs

What problem does IAIML address in interpretable tabular classifiers?

IAIML targets the weakness of marginal feature screening, which can remove variables that only become predictive through interactions with other variables.

How does IAIML detect and incorporate pairwise interactions?

IAIML uses adaptive per-feature discretization, finite-grid pairwise interaction scoring, and a partitioned explanation budget to admit and route detected interactions into a sparse downstream classifier.

What performance behavior does IAIML show across datasets?

On a 40-dataset panel, IAIML achieves AUC close to tuned gradient-boosted ensembles, while using roughly 14–28 times fewer fitted explanation components, and it performs best when pairwise interaction structure is strong with low marginal signal.

Sep 21, 2026

Uploaded by Valentina

Complexity-Budgeted, Interaction-Aware Interpretable Model for Tabular Data - Abstract & Introduction_Page 1

Open in WPS to view all 31 pages