Numerical Association Rule Mining from a Defined Schema Using the VMO Algorithm

Iván Fredy Jaramillo,Javier Garzás,Andrés Redchuk

doi:10.3390/app11136154

Iván Fredy Jaramillo, Javier Garzás + Show 1 more

Open Access

https://doi.org/10.3390/app11136154

Copy DOI

Journal: Applied Sciences	Publication Date: Jul 2, 2021
Citations: 2	License type: CC BY 4.0

Affiliation: King Juan Carlos University

Abstract

Association rule mining has been studied from various perspectives, all of which have made valuable contributions to data science. However, there are promising research lines, such as the inclusion of continuous variables and the combination of numerical and categorical attributes for a supervised classification variety. This research presents a new alternative for solving the numerical association rule-mining problem from an optimization perspective by using the VMO (Variable Mesh Optimization) meta-heuristic. This work includes the ability for classification when categorical data are available from a defined rule schema. Our technique implements an optimization process for the intervals of continuous variables, unlike others that discretize these types of variables. Some experiments were carried out with a real dataset to evaluate the quality of the rules obtained; in addition to this, this technique was compared with four population-based algorithms. The results show that this implementation is competitive in classification cases and has more satisfactory results for completely numerical data.

Highlights

One of the most significant contributions to data science is that of association rule mining (ARM) [2]; from its origin to the present, it has been applied to the discovery of purchasing patterns by starting with transactional data
This study resulted in a version of the Variable Mesh Optimization (VMO) algorithm used to solve the quantitative association rule mining problem
We developed QM_VMO, which is an implementation based on the VMO algorithm, in order to solve the problem of the extraction of numerical association rules

Summary

Introduction

It is known that the datasets are not composed only of qualitative attributes, and quantitative ones, which have different characteristics, such as the wide range in which the values are defined. They do not occur with a significant frequency in the set, making it difficult to identify patterns with techniques similar to those used with categorical attributes. We refer to quantitative association rule mining (QARM) [7]. This is a subdivision that encompasses numerical attributes in the search process for patterns through rules in the data

Methods

Results

Conclusion