AI2030 Responsible AI Tools Marketplace

Key Takeaways

KANs offer a solution to the "black box" problem in clinical AI: Kolmogorov-Arnold Networks provide interpretable alternatives to conventional machine learning models while maintaining predictive performance comparable to or exceeding standard baselines, addressing a critical barrier to clinical adoption where predictions must be auditable and actionable.
Two complementary approaches balance performance and transparency: Logistic KAN achieved the highest overall ranking (mean reciprocal rank of 0.76) across evaluation metrics, while KAAM sacrifices minimal accuracy in exchange for enhanced transparency through feature-wise decomposability and patient-level visualizations that enable direct inspection of individual predictions.
Feature-wise decomposability enables clinically meaningful explanations: KAAM's additive structure allows healthcare providers to understand how individual patient characteristics contribute to predictions, and supports nearest-patient retrieval—allowing clinicians to compare similar cases—rather than relying on abstract numerical explanations.
Interpretability doesn't require sacrificing performance: Across multiple public clinical datasets, both KAN-based models achieved competitive or superior accuracy compared to linear models, tree-based methods, and neural network baselines, suggesting that transparency and predictive power are not inherently at odds in clinical classification tasks.
Practical implementation is feasible for clinical settings: The models generate symbolic representations and patient-level visualizations suitable for clinical workflows, moving beyond theoretical interpretability to provide actionable insights that clinicians can directly use in decision-making.

Abstract

The increasing use of machine learning in clinical decision support has been limited by the lack of transparency of many high-performing models. In clinical settings, predictions must be interpretable, auditable, and actionable. This study investigates Kolmogorov-Arnold Networks (KANs) as intrinsically interpretable alternatives to conventional black-box models for clinical classification of tabular health data, aiming to balance predictive performance with clinically meaningful transparency. We introduce two KAN-based models: the Logistic KAN, a flexible generalization of logistic regression, and the Kolmogorov-Arnold Additive Model (KAAM), an additive variant that yields transparent symbolic representations through feature-wise decomposability. Both models are evaluated on multiple public clinical datasets and compared with standard linear, tree-based, and neural baselines. Across all datasets, the proposed models achieve predictive performance comparable to or exceeding that of commonly used baselines while remaining fully interpretable. Logistic-KAN obtains the highest overall ranking across evaluation metrics, with a mean reciprocal rank of 0.76, indicating consistently strong performance across tasks. KAAM provides competitive accuracy while offering enhanced transparency through feature-wise decomposability, patient-level visualizations, and nearest-patient retrieval, enabling direct inspection of individual predictions. KAN-based models provide a practical and trustworthy alternative to black-box models for clinical classification, offering a strong balance between predictive performance and interpretability for clinical decision support. By enabling transparent, patient-level reasoning and clinically actionable insights, the proposed models represent a promising step toward trustworthy AI in healthcare (code: https://github.com/Patricia-A-Apellaniz/classification_with_kans).

29 pages

Key Takeaways

KANs offer a solution to the "black box" problem in clinical AI: Kolmogorov-Arnold Networks provide interpretable alternatives to conventional machine learning models while maintaining predictive performance comparable to or exceeding standard baselines, addressing a critical barrier to clinical adoption where predictions must be auditable and actionable.
Two complementary approaches balance performance and transparency: Logistic KAN achieved the highest overall ranking (mean reciprocal rank of 0.76) across evaluation metrics, while KAAM sacrifices minimal accuracy in exchange for enhanced transparency through feature-wise decomposability and patient-level visualizations that enable direct inspection of individual predictions.
Feature-wise decomposability enables clinically meaningful explanations: KAAM's additive structure allows healthcare providers to understand how individual patient characteristics contribute to predictions, and supports nearest-patient retrieval—allowing clinicians to compare similar cases—rather than relying on abstract numerical explanations.
Interpretability doesn't require sacrificing performance: Across multiple public clinical datasets, both KAN-based models achieved competitive or superior accuracy compared to linear models, tree-based methods, and neural network baselines, suggesting that transparency and predictive power are not inherently at odds in clinical classification tasks.
Practical implementation is feasible for clinical settings: The models generate symbolic representations and patient-level visualizations suitable for clinical workflows, moving beyond theoretical interpretability to provide actionable insights that clinicians can directly use in decision-making.

Abstract

29 pages

Interpretable Clinical Classification with Kolmogorov-Arnold Networks

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links

Interpretable Clinical Classification with Kolmogorov-Arnold Networks

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links