|Online Discovery of Motifs in Time Series|
KMOTIF monitors time series data and discovers meaningful motifs in an online manner.
Below, you can find some extra information:
|KMOTIF finds the top-k most similar motifs in time series data in an online manner. A motif in time series data is the repeated sequences. For example, the following picture shows an example of top-2 motifs discovered in the brain activity data:|
Figure 1: An example of top-2 motifs in brain activity data. Motif are subsequences with a similar shape.
|Our main research results can be summarized as follows:
The comparison of kMotif and nMotif to the existing algorithm by Mueen and Keogh oMotif is illustrated in the following table:
|In brain activity a motif may be in the form
of the largest connected components (brain areas) preceding a seizure.
Therefore, finding the significant motifs can be valuable for predicting
the seizure periods. In other words, the strong similarity in motif
occurrences in seizure data not only can be used in forecasting the
unobserved outcome but also can help to isolate groups of neurons that
trigger identical activity during the epileptic attacks [Sauer et
In order to identify significant motifs we need powerful visualization tools which show the similarity structure between the discovered motifs. For instance, we carried out a simple experiment with the brain activity datasets from epileptic patients during epileptic attacks [Yankov et al.]. This dataset consists of 100 time series of length 4087 each. The motif length m is set to 174 (the number of recordings per sec.) and the sliding window length is set to w = 4087.
Figure 2: Motifs discovered by kMoif algorithm in the brain activity data
|Download the code|
|The source code was carefully cross checked by the authors of the work. You can download the code here. Please kindly send us an email for the password to extract the Zip files. If you have any question regarding the usage and the possible bugs that you discover in the code, please also contact us for possible assistant.|
|Download the data|
|The datasets we use in our work including EOG, EEG, Insect, Random Walk and the Brain Activity all can be downloaded on A. Mueen's websites|
|You can download the paper here.|
|We would like to thank NWO for their generous funding for our COMPASS project. We deeply thank A. Mueen and professor E. Keogh for their released datasets, source code and useful discussion in the early stage of the project. We also thank all the anonymous reviewers for their useful comments which help us improve our work significantly.|
Copyright TU/e 2010 by Hoang Thanh Lam, Toon Calders and Pham Dang Ninh