Abstract
Time series data are abundant in various domains and are often characterized as large in size and high in dimensionality, leading to storage and processing challenges. Symbolic representation of time series-which transforms numeric time series data into texts-is a promising technique to address these challenges. However, these techniques are essentially lossy compression functions and information are partially lost during transformation. To that end, we bring up a novel approach named Domain Series Corpus (DSCo), which builds per-class language models from the symbolized texts. To classify unlabeled samples, we compute the fitness of each symbolized sample against all per-class models and choose the class represented by the model with the best fitness score. Our work innovatively takes advantage of mature techniques from both time series mining and NLP communities. Through extensive experiments on an open dataset archive, we demonstrate that it performs similarly to approaches working with original uncompressed numeric data.
| Original language | English |
|---|---|
| Title of host publication | Machine Learning and Data Mining in Pattern Recognition - 12th International Conference, MLDM 2016, Proceedings |
| Publisher | Springer-Verlag London Ltd. |
| Pages | 294-310 |
| Number of pages | 17 |
| Volume | 9729 |
| ISBN (Print) | 9783319419190 |
| DOIs | |
| Publication status | Published - 2016 |
| Externally published | Yes |
| Event | 12th International Conference on Machine Learning and Data Mining in Pattern Recognition, MLDM 2016 - New York, United States of America Duration: 16 Jul 2016 → 21 Jul 2016 |
Publication series
| Name | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
|---|---|
| Volume | 9729 |
| ISSN (Print) | 0302-9743 |
| ISSN (Electronic) | 1611-3349 |
Conference
| Conference | 12th International Conference on Machine Learning and Data Mining in Pattern Recognition, MLDM 2016 |
|---|---|
| Country | United States of America |
| City | New York |
| Period | 16/07/16 → 21/07/16 |
Cite this
}
DSCo : A language modeling approach for time series classification. / Li, Daoyuan; Li, Li; Bissyandé, Tegawendé F.; Klein, Jacques; Traon, Yves Le.
Machine Learning and Data Mining in Pattern Recognition - 12th International Conference, MLDM 2016, Proceedings. Vol. 9729 Springer-Verlag London Ltd., 2016. p. 294-310 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 9729).Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › Research › peer-review
TY - GEN
T1 - DSCo
T2 - A language modeling approach for time series classification
AU - Li, Daoyuan
AU - Li, Li
AU - Bissyandé, Tegawendé F.
AU - Klein, Jacques
AU - Traon, Yves Le
PY - 2016
Y1 - 2016
N2 - Time series data are abundant in various domains and are often characterized as large in size and high in dimensionality, leading to storage and processing challenges. Symbolic representation of time series-which transforms numeric time series data into texts-is a promising technique to address these challenges. However, these techniques are essentially lossy compression functions and information are partially lost during transformation. To that end, we bring up a novel approach named Domain Series Corpus (DSCo), which builds per-class language models from the symbolized texts. To classify unlabeled samples, we compute the fitness of each symbolized sample against all per-class models and choose the class represented by the model with the best fitness score. Our work innovatively takes advantage of mature techniques from both time series mining and NLP communities. Through extensive experiments on an open dataset archive, we demonstrate that it performs similarly to approaches working with original uncompressed numeric data.
AB - Time series data are abundant in various domains and are often characterized as large in size and high in dimensionality, leading to storage and processing challenges. Symbolic representation of time series-which transforms numeric time series data into texts-is a promising technique to address these challenges. However, these techniques are essentially lossy compression functions and information are partially lost during transformation. To that end, we bring up a novel approach named Domain Series Corpus (DSCo), which builds per-class language models from the symbolized texts. To classify unlabeled samples, we compute the fitness of each symbolized sample against all per-class models and choose the class represented by the model with the best fitness score. Our work innovatively takes advantage of mature techniques from both time series mining and NLP communities. Through extensive experiments on an open dataset archive, we demonstrate that it performs similarly to approaches working with original uncompressed numeric data.
UR - http://www.scopus.com/inward/record.url?scp=84978976230&partnerID=8YFLogxK
U2 - 10.1007/978-3-319-41920-6_22
DO - 10.1007/978-3-319-41920-6_22
M3 - Conference Paper
SN - 9783319419190
VL - 9729
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 294
EP - 310
BT - Machine Learning and Data Mining in Pattern Recognition - 12th International Conference, MLDM 2016, Proceedings
PB - Springer-Verlag London Ltd.
ER -