A Developed Algorithm of Apriori Based on Association Analysis

Based on association analysis , an improved algorithm of Apriori is presented in the paper. The main idea of the algorithm are: (1) Count the probability of each attribute item(A1 , A2,...Am) of a DB by scanning the DB first time; (2)The probability of any two items Ak and Am appeared synchronously in one record is Pkm. min( Pk , Pm )≤Pkm ≤Pk *Pm , if Ak and Am is total correlation, then the Pkm is the minimum of the Pk and Pm,; if Ak and Am is total independent, then the Pkm is Pk *Pm; So we can estimate : Pkm =(a*min(Pk, Pm)+b*Pk*Pm)/(a+b); a+b=1 Parameter “a” is the probability while Ak and Am are total correlation, Parameter “b” is the probability while Ak and Am are total independent, Parameter “a” and “b” can use other method such as association analysis to count. In this paper a method for calculate the parameter “a” and ”b” with association analysis is provided. if Pkm is more than the threshold value which the user set, then Ak , Am are the frequent itemsets. You can use the method which described above to find out all the frequent itemsets without scanning DB so many times. (3)Count the support of the frequent itemsets by scanning the DB another time; (4)Output the association rules from the frequent itemsets. The detailed algorithm and it's sample are described in the paper . At last we compared it with algorithm apriori. The best quality is that the algorithm in our paper reduce the times of scanning DB.