With the explosive growth of data, mining information and knowledge from large databases has become one of the major challenges for data management and mining community. Data mining is the extraction of hidden unpredictive information from large databases. It is concerned with the analysis of data and finding patterns that exist in large databases but are hidden among the vast amount of data. Association rules are one of the most popular data mining techniques. The first step in mining association rules is mining frequent patterns. They are particularly useful for discovering relationships among data in huge databases. This study proposes a novel scheme for candidate generation that generates all the candidate item sets in three iterations. A new algorithm called AR-mine for association rule mining is also presented that uses the proposed scheme for candidate generation. A distinct feature of this algorithm is that a candidate item set is generated only when it actually encounters an occurrence of that set in the database. Another important feature is that it requires only three scans of the database. A simple hash table is used to store the candidate item sets, which speeds up the searching process. Our experiments with synthetic data sets and real life data sets show that AR-mine performs better than apriori, a well known and widely used algorithm for association rule mining.
展开▼
