Variable length motif discovery in time series data

The detection of recurring behavioral patterns in time series data, also called motif discovery, is a crucial step for mining insights complex especially environments where manual monitoring not feasible. However, current state-of-the-art algorithms fall short their applicability production (due to static length, lots user defined parameters, only providing the best pair, etc.). In this paper, variable length discovery method proposed based on Matrix Profile which focuses industrial applicability. It works noisy and periodic environments, returns unique motifs (meaning same shape are grouped together as one) requires one distance matrix calculation. was benchmarked synthetic data well publicly available real world key performance indicator (KPI) from telecom providers shows adequate accuracy finding both long series.