Skip to Main content Skip to Navigation

Knowledge extraction from uncertain and cyclic time series : application to Manual Wheelchair locomotion analysis

Abstract : This thesis addresses scientific issues from a data science perspective as part of the analysis of time series from manual wheelchair locomotion (FRM).Compression and classification with Dynamic Time Warping: Dyna- mic Time Warping (DTW) is a time series alignment algorithm that is often used because it considers that it exits small distortions between time series during their alignment. However, DTW sometimes produces pathological alignments that occur when, during the comparison of two time series X and Y, one data point of the time series X is compared to a large subsequence of data points of Y. In this chapter, we demonstrate that compressing time series using Piecewise Aggregate Approximation (PAA) is a simple strategy that greatly increases the quality of the alignment with DTW. This result is particularly true for synthetic data sets.Frobenius correlation based u-shapelets discovery for time series clustering: An unsupervised shapelet (u-shapelet) is a sub-sequence of a time series used for clustering a time series dataset. The purpose of this chapter is to discover u-shapelets on uncertain time series. To achieve this goal, we propose a dissimilarity score robust to uncertainty called FOTS whose computation is based on the eigen- vector decomposition and the comparison of the autocorrelation matrices of the time series. This score is robust to the presence of uncertainty; it is not very sensitive to transient changes; it allows capturing complex relationships between time series such as oscillations and trends, and it is also well adapted to the comparison of short time series. The FOTS score has been used with the Scalable Unsupervised Shapelet Discovery algorithm for the clustering of 17 datasets, and it has shown a substantial improvement in the quality of clustering with respect to the Rand Index. This work defines a novel framework for clustering of uncertain time series.Symbolic representation of cyclic time series based on properties of cycles: The analysis of cyclic time series from bio-mechanics is based on the comparison of the properties of their cycles. As usual algorithms of time series classification ignore this particularity, we propose a symbolic representation of cyclic time series based on the properties of cycles, named SAX-P. The resulting character strings can be compared using the Dynamic Time Warping distance. The application of SAX-P to propulsive moments of three subjects (S1, S2, S3) moving in Manual Wheelchair highlight the asymmetry of their propulsion. The symbolic representation SAX-P facilitates the reading of the cyclic time series and the clinical interpretation of the classification results.
Document type :
Complete list of metadatas

Cited literature [220 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Friday, September 20, 2019 - 10:32:21 AM
Last modification on : Wednesday, March 4, 2020 - 12:28:05 PM
Long-term archiving on: : Sunday, February 9, 2020 - 12:11:02 AM


Version validated by the jury (STAR)


  • HAL Id : tel-02292730, version 1



Vanel Steve Siyou Fotso. Knowledge extraction from uncertain and cyclic time series : application to Manual Wheelchair locomotion analysis. Other [cs.OH]. Université Clermont Auvergne, 2018. English. ⟨NNT : 2018CLFAC086⟩. ⟨tel-02292730⟩



Record views


Files downloads