Abstract

Detecting unusual or interesting patterns in discrete symbol sequences is of great importance. Many domains consist of discrete sequential time-series such as internet traffic, online transactions, cyber-attacks, financial transactions, biological transcription, intensive care data and social sciences data such as career trajectories or residential history. The sequences usually consist of discrete symbols that may form regular patterns or motifs. We use regular expressions to construct the longest repeating sequences and sub-sequences that compose them, we then define these as motifs (which may or may not represent novel patterns). The sequences are now composed of simpler motifs which are used to build Hidden Markov Models models which can capture complex relationships based on location, frequency of occurrence and position. New data that deviates from established motifs either in location of appearance, frequency of appearance, or motif composition may represent patterns that may be different in some way and hence interesting to the user.