Associating absent frequent itemsets with infrequent items to identify abnormal transactions

Data stored in transactional databases are vulnerable to noise and outliers and are often discarded at the early stage of data mining. Abnormal transactions in the marketing transactional database are those transactions that should contain some items but do not. However, some abnormal transactions may provide valuable information in the knowledge mining process. The literature on how to efficiently identify abnormal transactions in the database as well as determine what causes the transactions to be abnormal is scarce. This paper proposes a framework to realize abnormal transactions as well as the items that induce the abnormal transactions. Results from one synthetic and two medical data sets are presented to compare with previous work to verify the effectiveness of the proposed framework.