Mining Episode Rules from Event Sequences Under Non-overlapping Frequency

Abstract

Frequent episode mining is a popular framework for retrieving useful information from an event sequence. Many algorithms have been proposed to mine frequent episodes and to derive episode rules from them with respect to a given frequency function and its properties such as the anti-monotony. However, the interpretation of these rules is often difficult as their occurrences are allowed to overlap. To address this issue, this paper studies the novel problem of mining episode rules using non-overlapping occurrences of frequent episodes. The proposed rules have the form $\beta \Rightarrow \alpha$ where $\alpha$ and $\beta$ are frequent episodes and $\beta$ is a prefix of $\alpha$. This kind of rules is well adapted for prediction tasks where a phenomenon is predicted from some observed event(s). An efficient algorithm named NONEPI (NON overlapping EPIsode rule miner) is presented and experiments have been performed to compare its performance with state-of-the-art algorithms.