ABSTRACT

Efficient algorithms to discover frequent patterns are essential in data mining research. Frequent pattern mining is emerging as powerful tool for many business applications such as e-commerce, recommender systems and supply chain management and group decision support systems to name a few. Several effective data structures, such as two-dimensional arrays, graphs, trees and tries have been proposed to collect candidate and frequent itemsets. It seems as the tree structure is most extractive to storing itemsets. The outstanding tree has been proposed so far is called FP-tree which is a prefix tree structure. Some advancement with the FP tree structure is proposed as CATS tree. CATS Tree extends the idea of FP-Tree to improve storage compression and allow frequent pattern mining without generation of candidate itemsets. It allows to mine only through a single pass over the database. The efficiency of Apriori, FP-Growth, CATS Tree for incremental mining is very poor. In all of the above mentioned algorithms, it is required to generate tree repeatedly to support incremental mining. The implemented CATSIM Tree uses more memory compared to Apriori, FP-Growth and CATS Tree, but with advancement in technology, is not a major concern. In this work CATSIM Tree with modifications in CATS Tree is implemented to support incremental mining with better results.