An Empirical Algorithm for High and Low Correlative Association Rule Mining

Abstract

Support and confidence based Association rules mining algorithms have certain problems. Although other metrics like interest factor, comprehensibility, lift, correlation etc. are available to measure the interestingness of association rules. All the objectives are not suitable for each and every situation. All the objectives which were proposed in the literature, have some drawback, like correlation analysis gives equal importance to the items those are present and absent in transaction database. Resultant the rules generated by this, sometime mislead decision makers. Hence there is a strong need to define some new objectives for association rules that support in effective decision making. In this paper, authors proposed two novel objectives, high correlation and low correlation for 2-variables and 3-variables. These novel objectives clearly indicate that how much or how less two/three items are correlated. On the basis of this, decision makers can form their business strategies. An empirical algorithm for high and low correlative association rules generation is also proposed. With numeric evolution and experiments on the real-life data set, effectiveness has been measured and found that proposed algorithm gives better results.