Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neurosymbolic Association Rule Mining from Tabular Data

Erkan Karabulut, Paul Groth, Victoria Degeler

TL;DR

This work tackles the rule explosion and computational bottlenecks in association rule mining for high-dimensional tabular data by introducing Aerial+, a neurosymbolic approach that uses an under-complete denoising autoencoder to learn feature associations and reconstruct-based rule extraction to generate concise, high-quality rules with full data coverage. The method enables scalable, GPU-accelerated rule mining and integrates effectively with rule-based interpretable ML models, reducing runtime while maintaining or improving accuracy in downstream tasks. Extensive experiments on five UCI datasets show Aerial+ outperforms baselines in rule conciseness and speed, and two proposed variations further extend applicability to item-constrained ARM and frequent itemset mining. Overall, Aerial+ demonstrates a practical, scalable pathway for bridging deep representation learning with symbolic rule mining to improve interpretability in high-stakes decision-making tasks.

Abstract

Association Rule Mining (ARM) is the task of mining patterns among data features in the form of logical rules, with applications across a myriad of domains. However, high-dimensional datasets often result in an excessive number of rules, increasing execution time and negatively impacting downstream task performance. Managing this rule explosion remains a central challenge in ARM research. To address this, we introduce Aerial+, a novel neurosymbolic ARM method. Aerial+ leverages an under-complete autoencoder to create a neural representation of the data, capturing associations between features. It extracts rules from this neural representation by exploiting the model's reconstruction mechanism. Extensive evaluations on five datasets against seven baselines demonstrate that Aerial+ achieves state-of-the-art results by learning more concise, high-quality rule sets with full data coverage. When integrated into rule-based interpretable machine learning models, Aerial+ significantly reduces execution time while maintaining or improving accuracy.

Neurosymbolic Association Rule Mining from Tabular Data

TL;DR

This work tackles the rule explosion and computational bottlenecks in association rule mining for high-dimensional tabular data by introducing Aerial+, a neurosymbolic approach that uses an under-complete denoising autoencoder to learn feature associations and reconstruct-based rule extraction to generate concise, high-quality rules with full data coverage. The method enables scalable, GPU-accelerated rule mining and integrates effectively with rule-based interpretable ML models, reducing runtime while maintaining or improving accuracy in downstream tasks. Extensive experiments on five UCI datasets show Aerial+ outperforms baselines in rule conciseness and speed, and two proposed variations further extend applicability to item-constrained ARM and frequent itemset mining. Overall, Aerial+ demonstrates a practical, scalable pathway for bridging deep representation learning with symbolic rule mining to improve interpretability in high-stakes decision-making tasks.

Abstract

Association Rule Mining (ARM) is the task of mining patterns among data features in the form of logical rules, with applications across a myriad of domains. However, high-dimensional datasets often result in an excessive number of rules, increasing execution time and negatively impacting downstream task performance. Managing this rule explosion remains a central challenge in ARM research. To address this, we introduce Aerial+, a novel neurosymbolic ARM method. Aerial+ leverages an under-complete autoencoder to create a neural representation of the data, capturing associations between features. It extracts rules from this neural representation by exploiting the model's reconstruction mechanism. Extensive evaluations on five datasets against seven baselines demonstrate that Aerial+ achieves state-of-the-art results by learning more concise, high-quality rule sets with full data coverage. When integrated into rule-based interpretable machine learning models, Aerial+ significantly reduces execution time while maintaining or improving accuracy.
Paper Structure (21 sections, 11 equations, 6 figures, 7 tables, 3 algorithms)

This paper contains 21 sections, 11 equations, 6 figures, 7 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Neurosymbolic Rule Mining Pipeline
  5. Autoencoder Architecture and Training Stage
  6. Rule Extraction Stage
  7. Evaluation
  8. Experimental Setting 1: Execution Time and Rule Quality Evaluation
  9. Experimental Setting 2: Aerial+ on Downstream Tasks
  10. Conclusions
  11. Runtime Complexity Analysis of Aerial+
  12. Hyperparameter Analysis of Aerial+
  13. Variations of Aerial+
  14. Execution Time Experiments for Optimization-based ARM
  15. Effect of Minimum Support Threshold on Classification
  16. ...and 6 more sections

Figures (6)

  • Figure 1: Neurosymbolic ARM pipeline of Aerial+.
  • Figure 2: Aerial+ rule extraction example.
  • Figure 3: Exhaustive methods incur higher execution times as antecedents increase (top) or support threshold decreases (bottom).
  • Figure 4: Aerial+ yields fewer rules and lower execution time than exhaustive methods as antecedents increase.
  • Figure 5: Increasing $\tau_a$ results in a lower number of rules with higher support.
  • ...and 1 more figures