Efficient mining of all margin-closed itemsets with applications in temporal knowledge discovery and classification by compression

Margin-closed itemsets have previously been proposed as a subset of the closed itemsets with a minimum margin constraint on the difference in support to supersets. The constraint reduces redundancy in the set of reported patterns favoring longer, more specific patterns. A variety of patterns ranging from rare specific itemsets to frequent general itemsets is reported to support exploratory data analysis and understandable classification models. We present DCI_Margin, a new efficient algorithm that mines the complete set of margin-closed itemsets. We modified the DCI_Closed algorithm that has low memory requirements and can be parallelized. The margin constraint is checked on-the-fly reusing information already computed by DCI_Closed. We thoroughly analyzed the behavior on many datasets and show how other data mining algorithms can benefit from the redundancy reduction.