## Segment and combine approach for non-parametric time-series classification (2005)

Venue: | in Proceedings of the 9th European Conference on Principles and Practice of Knowledge Discovery in Databases (PKDD |

Citations: | 7 - 4 self |

### BibTeX

@INPROCEEDINGS{Geurts05segmentand,

author = {Pierre Geurts and Louis Wehenkel},

title = {Segment and combine approach for non-parametric time-series classification},

booktitle = {in Proceedings of the 9th European Conference on Principles and Practice of Knowledge Discovery in Databases (PKDD},

year = {2005},

pages = {478--485}

}

### OpenURL

### Abstract

Abstract. This paper presents a novel, generic, scalable, autonomous, and flexible supervised learning algorithm for the classification of multivariate and variable length time series. The essential ingredients of the algorithm are randomization, segmentation of time-series, decision tree ensemble based learning of subseries classifiers, combination of subseries classification by voting, and cross-validation based temporal resolution adaptation. Experiments are carried out with this method on 10 synthetic and real-world datasets. They highlight the good behavior of the algorithm on a large diversity of problems. Our results are also highly competitive with existing approaches from the literature. 1 Learning to classify time-series Time-series classification is an important problem from the viewpoint of its multitudinous applications. Specific applications concern the non intrusive monitoring and diagnosis of processes and biological systems, for example to decide whether the system is in a healthy operating condition on the basis of measurements

### Citations

150 |
The UCI KDD archive
- Bay
- 1999
(Show Context)
Citation Context ...136 holdout 1000 2.10 [7] 1,2,5,10,20,30,40 JV 2 640 8 10 7-29 holdout 270 3.80 [8] 2,3,5,7 ECG 4 200 2 2 39-152 10-fold cv – 1,2,5,10,20,30,39 1 http://www.montefiore.ulg.ac.be/~geurts/thesis.html 2 =-=[5]-=- 3 http://www2.ife.no 4 http://www-2.cs.cmu.edu/~bobski/pubs/tr01108.html 5 http://waleed.web.cse.unsw.edu.au/new/phd.html described in details in [4]. It grows a tree by selecting the best split from... |

72 | L.: Random subwindows for robust image classification
- Marée, Geurts, et al.
(Show Context)
Citation Context ...ed that the method could be used for real-time time-series classification, by adjusting the voting scheme. The approach presented here for time-series is essentially identical to the work reported in =-=[9]-=- for image classification. Similar ideas could also be exploited to yield generic approaches for the classification of texts or biological sequences. Although these latter problems have different stru... |

60 | Making time-series classification more accurate using learned constraints
- Ratanamahatana, Keogh
- 2004
(Show Context)
Citation Context ...temporal specific peculiarities (e.g. invariance with respect tostime or amplitude rescaling) and then to use this distance measure in combination with nearest neighbors or other kernel-based methods =-=[12, 13]-=-. A potential advantage of these approaches is the possibility to bias the representation by exploiting prior problem specific knowledge. At the same time, this problem specific modeling step makes th... |

54 | Dynamic time-alignment kernel in support vector machine
- Shimodaira, Noma, et al.
- 2002
(Show Context)
Citation Context ...temporal specific peculiarities (e.g. invariance with respect tostime or amplitude rescaling) and then to use this distance measure in combination with nearest neighbors or other kernel-based methods =-=[12, 13]-=-. A potential advantage of these approaches is the possibility to bias the representation by exploiting prior problem specific knowledge. At the same time, this problem specific modeling step makes th... |

53 | Pattern Extraction for Time Series Classification
- Geurts
- 2001
(Show Context)
Citation Context ...merical) features which can then be used as input representation for any base learner (e.g. [7, 8, 10, 11]). This feature extraction step can also be incorporated directly into the learning algorithm =-=[1, 2, 14]-=-. Another approach is to define a distance or similarity measure between time-series that takes into account temporal specific peculiarities (e.g. invariance with respect tostime or amplitude rescalin... |

48 | Learning Comprehensible Descriptions of Multivariate Time Series
- Kadous
- 1999
(Show Context)
Citation Context ... 13.0 ± 4.6 Auslan-b 22.82 10 4.51 10 18.40 40.0 ± 0.0 5.16 40.0 ± 0.0 JV 16.49 2 4.59 2 8.11 3 4.05 3 ECG 25.00 18.5 ± 10.0 15.50 19.0 ± 9.4 25.50 29.8 ± 6.0 24.00 32.4 ± 8.5 normalization technique =-=[2, 6]-=-, which aims at transforming a time-series into a vector of fixed dimensionality of scalar numerical attributes: the time interval of each object is divided into s equal-length segments and the averag... |

30 | Automatic feature extraction for classifying audio data
- Mierswa, Morik
- 2005
(Show Context)
Citation Context ...ollection of temporal predicates which can be applied to each time-series in order to compute (logical or numerical) features which can then be used as input representation for any base learner (e.g. =-=[7, 8, 10, 11]-=-). This feature extraction step can also be incorporated directly into the learning algorithm [1, 2, 14]. Another approach is to define a distance or similarity measure between time-series that takes ... |

24 |
Contributions to Decision Tree Induction: Bias/Variance Tradeoff and Time Series Classification
- Geurts
- 2002
(Show Context)
Citation Context ... and combine method, which is due to the virtual increase of the learning sample size and the averaging step and somewhat mitigates the effect variance reduction techniques like ensemble methods (see =-=[3]-=- for a discussion of bias and variance of the segment and combine method). From the values of ℓ ∗ in the last column of Table 2, it is clear that the optimal ℓ ∗ is a problem dependent parameter. Inde... |

22 | Generalized Feature Extraction for Structural Pattern Recognition in Time-Series Data
- Olszewski
- 2001
(Show Context)
Citation Context ...ollection of temporal predicates which can be applied to each time-series in order to compute (logical or numerical) features which can then be used as input representation for any base learner (e.g. =-=[7, 8, 10, 11]-=-). This feature extraction step can also be incorporated directly into the learning algorithm [1, 2, 14]. Another approach is to define a distance or similarity measure between time-series that takes ... |

17 |
Multidimensional Curve Classification Using Passing-Through Regions
- Kudo, Toyama, et al.
- 1999
(Show Context)
Citation Context ...ollection of temporal predicates which can be applied to each time-series in order to compute (logical or numerical) features which can then be used as input representation for any base learner (e.g. =-=[7, 8, 10, 11]-=-). This feature extraction step can also be incorporated directly into the learning algorithm [1, 2, 14]. Another approach is to define a distance or similarity measure between time-series that takes ... |

11 |
Classification of Multivariate Time Series and Structured Data Using Constructive Induction
- Kadous, Sammut
- 2005
(Show Context)
Citation Context |

11 | Decision-tree Induction from Time-series Data Based on aStandard-example Split Test
- Yamada, Suzuki, et al.
- 2003
(Show Context)
Citation Context ...merical) features which can then be used as input representation for any base learner (e.g. [7, 8, 10, 11]). This feature extraction step can also be incorporated directly into the learning algorithm =-=[1, 2, 14]-=-. Another approach is to define a distance or similarity measure between time-series that takes into account temporal specific peculiarities (e.g. invariance with respect tostime or amplitude rescalin... |

1 |
Rodríguez Diez. Boosting interval-based literals: Variable length and early classification
- González, J
- 2004
(Show Context)
Citation Context ...merical) features which can then be used as input representation for any base learner (e.g. [7, 8, 10, 11]). This feature extraction step can also be incorporated directly into the learning algorithm =-=[1, 2, 14]-=-. Another approach is to define a distance or similarity measure between time-series that takes into account temporal specific peculiarities (e.g. invariance with respect tostime or amplitude rescalin... |