NiaAutoARM: Automated Framework for Constructing and Evaluating Association Rule Mining Pipelines
Year: 2025
Type: article
Abstract: Numerical Association Rule Mining (NARM), which simultaneously handles both numerical and categorical attributes, is a powerful approach for uncovering meaningful associations in heterogeneous datasets. However, designing effective NARM solutions is a complex task involving multiple sequential steps, such as data preprocessing, algorithm selection, hyper-parameter tuning, and the definition of rule quality metrics, which together form a complete processing pipeline. In this paper, we introduce NiaAutoARM, a novel Automated Machine Learning (AutoML) framework that leverages stochastic population-based metaheuristics to automatically construct full association rule mining pipelines. Extensive experimental evaluation on ten benchmark datasets demonstrated that NiaAutoARM consistently identifies high-quality pipelines, improving both rule accuracy and interpretability compared to baseline configurations. Furthermore, NiaAutoARM achieves superior or comparable performance to the state-of-the-art VARDE algorithm while offering greater flexibility and automation. These results highlight the framework’s practical value for automating NARM tasks, reducing the need for manual tuning, and enabling broader adoption of association rule mining in real-world applications.
Source: Mathematics
DOI: 10.3390/math13121957
Authors: Uroš Mlakar, Iztok Fister, Iztok Fister
Institution: University of Maribor
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,137
OpenAlex ID: W4411278151