PRESEE: An MDL/MML Algorithm to Time-Series Stream Segmenting

摘要: Time-series stream is one of the most common data types in mining field. It prevalent fields such as stock market, ecology, and medical care. Segmentation a key step to accelerate processing speed time-series mining. Previous algorithms for segmenting mainly focused on issue ameliorating precision instead paying much attention efficiency. Moreover, performance these depends heavily parameters, which are hard users set. In this paper, we propose PRESEE (parameter-free, real-time, scalable algorithm), greatly improves efficiency segmenting. based both MDL (minimum description length) MML message methods, could segment automatically. To evaluate PRESEE, conduct several experiments streams different compare it with state-of-art algorithm. The empirical results show that very efficient real-time datasets by improving nearly ten times. novelty algorithm further demonstrated application from ChinaFLUX sensor networks stream.