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 |
| Subtitle of host publication | 12th International Conference, MLDM 2016 New York, NY, USA, July 16–21, 2016 Proceedings |
| Editors | Petra Perner |
| Place of Publication | Cham Switzerland |
| Publisher | Springer |
| Pages | 294-310 |
| Number of pages | 17 |
| ISBN (Electronic) | 9783319419206 |
| ISBN (Print) | 9783319419190 |
| DOIs | |
| Publication status | Published - 2016 |
| Externally published | Yes |
| Event | International Conference on Machine Learning and Data Mining in Pattern Recognition 2016 - New York, United States of America Duration: 16 Jul 2016 → 21 Jul 2016 Conference number: 12th https://web.archive.org/web/20160304204738/http://www.mldm.de/ |
Publication series
| Name | Lecture Notes in Computer Science |
|---|---|
| Publisher | Springer |
| Volume | 9729 |
| ISSN (Print) | 0302-9743 |
| ISSN (Electronic) | 1611-3349 |
Conference
| Conference | International Conference on Machine Learning and Data Mining in Pattern Recognition 2016 |
|---|---|
| Abbreviated title | MLDM 2016 |
| Country | United States of America |
| City | New York |
| Period | 16/07/16 → 21/07/16 |
| Internet address |
Keywords
- Language Model
- Time Series Data
- Symbolic Representation
- Dynamic Time Warping
- Alphabet Size
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 New York, NY, USA, July 16–21, 2016 Proceedings. ed. / Petra Perner. Cham Switzerland : Springer, 2016. p. 294-310 (Lecture Notes in Computer Science ; Vol. 9729).Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › Research
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.
KW - Language Model
KW - Time Series Data
KW - Symbolic Representation
KW - Dynamic Time Warping
KW - Alphabet Size
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
AN - SCOPUS:84978976230
SN - 9783319419190
T3 - Lecture Notes in Computer Science
SP - 294
EP - 310
BT - Machine Learning and Data Mining in Pattern Recognition
A2 - Perner, Petra
PB - Springer
CY - Cham Switzerland
ER -