Correlation as an ARM Interestingness Measure for Numeric Datasets
Konstantinos Kelesidis, Nikoletta Fotopoulou, Dimitris A. Dervos · 2023
A key issue in Associating Rule Mining (ARM) is the handling of the huge number of association rules that emerge in the output of the data mining process. A number of interestingness measures have been proposed and are used in order to quantify the usefulness (relevance) of each one rule. This study approaches measure effectiveness in terms of mitigating information loss introduced by the noise inherent to the application considered, and/or by the data preparation stage. The approach is put to test by applying ARM on two numeric datasets: (a) the 1M MovieLens dataset, and (b) an academic dataset of student assessment scores accumulated over a twelve years period. The Pearson and Spearman correlation coefficients are used to rank the rules in the output of the ARM process. The results obtained indicate that the proposed approach leads to improved performance and provides more insight into the rules ranking problem when ARM is applied on collections involving numeric data of the ratio measurement scale.