OOIMASP: Origin based association rule mining with order independent mostly associated sequential patterns

Efficient mining of association rules on a transaction dataset is an interesting and a challenging problem. The state-of-the-art MASP algorithm is dependent on the order of items in the transaction. We propose OOIMASP algorithm, which has two novel properties- 1) order independence and 2) it takes into consideration the origin of items to calculate unbiased support and unbiased confidence values. Order dependence is one of the drawbacks of MASP. OOIMASP addresses this issue by rearranging the items in transactions using a greedy frequency based approach. We compare the performance of our system with MASP on five synthetic data sets and three public data sets. The results show that our proposed approach outperforms the MASP in both the comparison metrics, i.e., the number of association rules generated and the length of the longest association rule. Both these metrics are important to evaluate the performance of an algorithm. On an average, OOIMASP algorithm generates 632% longer rules and 457% more association rules than MASP algorithm. The disadvantage of the proposed algorithm is, it requires more computational resources in terms of time, approximately 5 times more than MASP. We claim that the extra information extracted using our method compensates for the increase in time complexity as compared to MASP. The proposed method produces multiple trees which can be very useful in the visual analysis of data.