A new equivalence class based approach for discretizing quantitative data using Point Shift Mechanism

Different data mining techniques are utilized for knowledge discovery in modern real-world databases. Such techniques quite often require discretization of values. The task of discretization of items in a database is basically, a kind of transformation of continuous values to a simpler form of comprehension. Discretization has far reaching effects on the entire knowledge discovery process and interestingly, at certain times these effects may be adverse. Association Rule Mining (ARM) in quantitative databases exploit discretization of quantitative attributes to generate reasonable intervals for mining meaningful association rules. But, there are several issues related to interval generation that originate due to inefficient discretization which ultimately degrade the quality of discovered association rules. In this paper, we discuss such issues inherent in some of the existing discretization techniques and highlight how they may affect the task of ARM in quantitative databases. We also propose a new discretization technique that tries to nullify the adverse effects of inefficient discretization. Experimental results reveal that the proposed technique outperforms several existing ones in terms of reasonable interval generation.