Abstract: During the search for frequent itemsets of the Apriori algorithm, the database is scanned repetitively and generates a large number of useless candidate sets. For this problem, a kind of improved Apriori algorithm based on the matrix reduction is put forward. The algorithm scans the database only once, converts the database information to Boolean matrix, and reduces the data structure according to the conclusion drawn from the nature of the frequent k-itemsets, which lowers the generation scale of the invalid candidate itemsets effectively. By comparing with the existing algorithms, it is validated that this algorithm can improve the efficiency of mining frequent itemsets effectively.

Key words: data mining, association rules, Apriori algorithm, frequent itemsets, matrix reduction