Mining itemsets in the presence of missing values

Missing values make up an important and unavoidable problem in data management and analysis. In the context of association rule and frequent itemset mining, however, this issue never received much attention. Nevertheless, the well known measures of support and confidence are misleading when missing values occur in the data, and more suitable definitions typically don't have the crucial monotonicity property of support. In this paper, we overcome this problem and provide an efficient algorithm, XMiner, for mining association rules and frequent itemsets in databases with missing values. XMiner is empirically evaluated, showing a clear gain over a straightforward baseline-algorithm.