A Graph-based Approach for Comparing Interestingness Measures
Hiep Xuan Huynh, Fabrice Guillet, H. Briand · 2006
In the context of data mining, we use the Spearman's rank correlation coefficient in order to compare the behavior of 40 interestingness measures of association rules. Via a new graph-based approach, we can visualize not only the strong but also the weak correlations between interestingness measures. We propose to discover the stable clusters of interestingness measures (i.e. subsets of interestingness measures delivering a close rule ranking) by making comparative study on two opposite datasets (a highly correlated one and a lowly correlated one). The results show that the correlation between interestingness measures depends on data nature and rule ranks, and show also 6 stable clusters.