Evaluation of the Serial Association Rule Mining Algorithms

F. Kovács, R. Iváncsy, and I. Vajk (Hungary)


data mining, association rule, level wise algorithms


One of the most important problems in data mining is to find association rules. The association rule mining algorithms can be classified into two main groups: the level-wise algorithms and the tree-based algorithms. The level-wise algorithms scan the entire database multiple time but they have moderate memory requirement. The two phase algorithms scan the database only twice but they can have extremely large memory requirement. During the evaluation of the level wise algorithms it is possible to realize that the first and the second database scan take considerable part of the total response time. In this paper a new level wise-algorithm is introduced, which reduce the costs of the first and the second database scan. The algorithm uses a multidimensional dynamic array data structure to reduce the candidate searching and item counting costs. This data structure has acceptable memory requirement and it accelerates substantially the item counting task. The novel algorithm can reach the performance of the tree-based algorithms and it has smaller memory requirement than its tree-based counterparts.

Important Links:

Go Back