Mining both frequent and rare episodes in multiple data streams

In this paper, we describe a method for mining both frequent episodes and rare episodes in multiple data streams. The main issues include episodes mining and data streams relationship processing. Therefore, a mining algorithm together with two dedicated handling mechanisms is presented. We propose the concept of alternative support for discovering frequent and rare episodes, and define the semantic similarity of event sequences for analyzing the relationships between data streams. The algorithm extracts basic episode information from each data stream and keeps the information in episode sets. Then analyze relationships of episode sets and merge similar episode sets, and mining episode rules from the merged sets by alternative support and confidence. From experiments, we find that our mining algorithm is successful for processing multiple data streams and mining frequent and rare episodes. Our research results may lead to a feasible solution for frequent and rare episodes mining in multiple data streams.