Constructing a new algorithm for high average utility Itemsets mining

Abstract

In this paper, we propose a mining algorithm for average-utility itemsets (EHAUI-Tree) based on improving HUUI-Tree algorithm to apply for adding new database transactions without restart. At first, the value of updated data is calculated. Then, itemsets which make changes will be calculated and updated depending upon the updated data value and the previous High Average-utility Upper-bound (HAUUB). This algorithm uses the downward closure property of an average-utility itemset and an index table structure. In addition, a data structure for itemsets is proposed to minimize memory usage and maximize calculating efficiency. The experimental result shows that EHAUI-Tree is more effective than HAUI-Tree when adding new transactions for the previous database. The method applies the downward closure properties of HAUUB Itemset and Index Table. Furthermore, the Bit-Array-structure itemset is also proposed to reduce using memory and calculate more effectively. The result of this algorithm is better than HAUI-Tree on updating new transactions.