Towards Associative Classification in Distributed Environments
(2026) European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases — Location: Naples, Italy (7.September.2026)
Files
2026_DistributedAC_ECMLPKDD.pdf
Open Access - Adobe PDF
- 638.84 KB
Details
- Authors
- Abstract
- We study associative classification (AC) in a fully trusted federated environment with horizontally partitioned data. We propose two distributed AC frameworks that enable multiple sites to collab-oratively train a global classifier while keeping raw data local. Both methods preserve the conventional AC pipeline at the local level (rule generation, ranking, and coverage-based pruning) while differing in how global rule metrics are computed. The first approach approximates global metrics using a communication-efficient consolidation strategy inspired by the DAC algorithm. The second computes exact global rule metrics via additional communication rounds, yielding classifiers that accurately approximate centralized models. To improve efficiency and reduce redundancy , the frameworks incorporate closed frequent itemset mining during rule generation. We evaluate the proposed methods on several datasets and compare them with DAC, logistic regression, and MLP. Exp erimen-tal results show that our refinements to the DAC framework substantially improve its predictive performance. Moreover, the exact-metric variant consistently yields accurate and stable classifiers across varying degrees of data distribution while maintaining relatively low communication costs. Overall, our findings suggest that high-quality associative classifiers can be constructed in distributed settings with minimal communication , offering a transparent and competitive alternative to black-box models.
- Affiliations
Citations
Veroneze, R., Lessage, X., & Vanderdonckt, J. (2026). Towards Associative Classification in Distributed Environments. Machine Learning and Knowledge Discovery in Databases, 2, 40-58. https://doi.org/10.1007/978-3-032-37657-2_3 (Original work published 2026)