Mining Partially-Ordered Episode Rules with the Head Support

Episode rule mining is a popular data analysis task that aims at finding rules describing strong relationships between events (or symbols) in a sequence. Finding episode rules can help understanding the data or making predictions. However, traditional episode rule mining algorithms find rules that require a very strict ordering between events. To loosen this ordering constraints and find more general and flexible rules, this paper presents an algorithm named POERMH (Partially-Ordered Episode Rule Mining with Head Support). Unlike previous algorithms, the head support frequency measure is used to select interesting episode rules. Experiments on real data show that POERMH can find interesting rules that also provides a good accuracy for sequence prediction.