Research And Application Of Incremental Updating Algorithm For Mining Association Rules

As an important field in data mining, the association rules mining presently has applied successfully in the fields of commerce, education, scientific research etc., which has become a most important and most active branch in data mining. Mining algorithms for association rules have been researched extensively, they are optimized and improved continuously. Whereas the efficiency of these algorithms is enhanced, there remain some deficiencies. In addition, there are two prevalent problems in mining association rules: How to acquire results efficiently and immediately when the database updates constantly? Usually, it's necessary to set some parameters for customers before mining, and mostly they have to adjust these parameters time after time to acquire the satisfactory rules, thus how to calculate efficiently during the repetitious process? Here, incremental updating algorithms can solve these problems. Based on the previous researches, we explore the incremental updating algorithms on mining association rules in this paper, the main content of which can be summed up as follows: Firstly, an introduction of some classic incremental updating algorithms and some optimized measures for mining association rules, and analysis of their advantages and disadvantages. Discussed and analyzed secondly are algorithm FUP and algorithm IUA with their improved algorithms, which are the characteristic algorithms in the case of adding new data to the database or the minimum support threshold lessening. By these researches, we provide two improved algorithms——MFUP and QIUA, and validate their feasibility and efficiency from theoretic and experimental aspects. Algorithm MFUP mainly focuses on the shortage in algorithm FUP that need scan database too many times. In algorithm MFUP, the times of scanning database is reduced greatly, when simultaneously, the pruning step can be overleaped. Algorithm QIUA mends defects in IUA which entails a mass of candidate sets by its join mode and may remove some potential frequent itemsets in pruning step, in which join mode will be improved and the mistake that removes some potential frequent itemsets can be corrected with the same pruning compare sets Thirdly, we apply the two algorithms to data mining in forestry. In this application, we analyze mining association rules in record-weighted data source before modifying the two algorithms to adapt this kind of data source and fulfill the mining task successfully.