Developing an Association Rules Mining Technique for Large Updated Databases

Abstract

Mining association rules is one of the important topics in the data mining field because it aims at discovering significant relationships between data items in input databases. With the massive amount of data supplied every day, the previously discovered relationships became invalid. The need of efficient and low-computationally cost algorithms is a big challenge currently for mining association rules in these large updated databases. In this thesis, we were interested in proposing new algorithms for solving incremental mining problem of large updated databases. In our study, we proposed IMIDB algorithm for discovering frequent itemsets from incremental traditional transactional databases. The IMIDB algorithm utilizes previously discovered frequent itemsets from original database and produces the frequent itemsets for the updated databases in efficient way compared to recently cited algorithms in this area. Another algorithm, IndxTAR, is also proposed for mining frequent itemsetes from incremental temporal databases. The proposed algorithm introduces a new and efficient indexing technique for indexing temporal databases that is used to facilitate the support counting process that was the bottleneck issue in existing incremental temporal mining algorithms. We were also interesting in investigating a new method for mining frequent weighted itemsets from weighted databases. So, wSWF algorithm is presented that adopts sliding window filtering technique to efficiently find all the frequent weighted itemsets. Chapter 6: Conclusions and Future Work 112 The algorithm performance was compared to recently cited algorithms and show great performance enhancement.