A Comparative Study of Tree-Based and Apriori-Based Approaches for Incremental Data Mining

Abstract

Association rule mining is an iterative and interactive process of discovering valid, novel, useful, understandable and hidden associations from the massive database. The Colossal databases require powerful and intelligent tools for analysis and discovery of frequent patterns and association rules. Several researchers have proposed the many algorithms for generating item sets and association rules for discovery of frequent patterns, and minning of the association rules. These proposals are validated on static data. A dynamic database may introduce some new association rules, which may be interesting and helpful in taking better business decisions. In association rule mining, the validation of performance and cost of the existing algorithms on incremental data are less explored. Hence, there is a strong need of comprehensive study and in-depth analysis of the existing proposals of association rule mining. In this paper, the existing tree-based algorithms for incremental data mining are presented and compared on the baisis of number of scans, structure, size and type of database. It is concluded that the Can-Tree approach dominates the other algorithms such as FP-Tree, FUFP-Tree, FELINE Alorithm with CATS-Tree etc.This study also highlights some hot issues and future research directions. This study also points out that there is a strong need for devising an efficient and new algorithm for incremental data mining.