Toward Reusing the Numerical Association Rule Mining Models

Abstract

Nowadays, it is no secret that modern machine learning methods are amongst the more computationally-intensive learning methods. The rise in the applications of computationally-intensive deep learning, automated machine learning methods, and even metaheuristics for optimization, have increased the consumption of electrical energy dramatically. Consequently, we can predict that the numbers of global carbon footprints, arising as a byproduct of the increased consumption of electrical energy during intensive computation, will be higher and higher in the near future. Fortunately, the research community is aware of these problems, and is, thus, looking for solutions of how to reduce the carbon footprint in the sense of the so-called Green AI. In line with this, the paper introduces a reusable model for Numerical Association Rule Mining which is also one of the hardest computationally-intensive learning methods. In the classical Numeric Association Rule Mining, the model (i.e., the archive of mined association rules) is created anew, when the new incoming association rule are emerging. The proposed reusable model only modifies the existing model accordingly by incoming new rules, while the complexity of computation decreases remarkably.