Semi-supervised time series classification

By Li Wei, Eamonn J. Keogh. Knowledge Discovery and Data Mining, 2006

304 citations recorded by the source corpus.

The problem of time series classification has attracted great interest in the last decade. However current research assumes the existence of large amounts of labeled training data. In reality, such data may be very difficult or expensive to obtain. For example, it may require the time and expertise of cardiologists, space launch technicians, or other domain specialists. As in many other domains, there are often copious amounts of unlabeled data available. For example, the PhysioBank archive contains gigabytes of ECG data. In this work we propose a semi-supervised technique for building time series classifiers. While such algorithms are well known in text domains, we will show that special considerations must be made to make them both efficient and effective for the time series domain. We evaluate our work with a comprehensive set of experiments on diverse data sources including electrocardiograms, handwritten documents, and video datasets. The experimental results demonstrate that our approach requires only a handful of labeled examples to construct accurate classifiers.

Subfields: Artificial Intelligence, Signal Processing

View Paper: https://www.semanticscholar.org/paper/807bfd050b9bd33d6dc8ff9533992d92a3590c5d

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Signal Processing

Citations in the corpus

Oboe: auto-tuning video ABR algorithms to network conditions (2018, methodology): “While several techniques exist for change point detection [22, 33, 36, 44, 54, 58], we focus on probabilistic methods [14, 18, 24, 57].”
Outlier Analysis (2013, methodology): “Semi-supervised methods for medical time-series classification are discussed in [462].”
A brief survey on sequence classification (2010, methodology): “Wei et al. [59] adopt one nearest neighbor classiﬁer for semi-supervised time series classiﬁcation.”
Rotation-invariant similarity in time series using bag-of-patterns representation (2012, background): “Classification of time series has attracted much interest from the data mining community (Geurts 2001; Keogh and Kasetty 2002; Nanopoulos et al. 2001; Ratanamahatana and Keogh 2004; Radovanovic et al. 2010; Wei and Keogh 2006; Ye and Keogh 2009; Mueen et al. 2011; Xing et al. 2011).”
Neural Networks for Sequential Data: a Pre-training Approach based on Hidden Markov Models (2015, unknown): “, [2]), editdistance (e.”
TapNet: Multivariate Time Series Classification with Attentional Prototypical Network (2020, methodology): “However, the issue of limited training labels can be handled by traditional time series classiﬁcation approach using distance-based methods such as DTW-1NN (Shokoohi-Yekta, Wang, and Keogh 2015).”
LSTM Fully Convolutional Networks for Time Series Classification (2017, background): “Time series data is ubiquitous [1], existing in weather readings [2], financial recordings [3], industrial observations [4], and psychological signals [5], [6].”
Highly Comparative Feature-Based Time-Series Classification (2014, methodology): “All the data is used exactly as obtained from the The UCR source [13], without any preprocessing and using the speciﬁed partitions of each dataset into training and test portions.”
A review on time series data mining (2011, methodology): “Ensembles are used to combine base classifiers, while Wei and Keogh (2006) study the combination of the numerosity reduction, using DTW and nearest-neighbor classifiers for time series classification.”
A Survey of Methods for Time Series Change Point Detection (2016, unknown): “Examples include decision tree [33][34][36][37], naïve Bayes [33], Bayesian net [34], support vector machine [33][34], nearest neighbor [33][20], hidden Markov model [38][39][33], conditional random field [34], and Gaussian mixture model (GMM) [38][39].”
TraClass: trajectory classification using hierarchical region-based and trajectory-based clustering (2008, methodology): “The 1-nearest neighbor classifler using dynamic time warping(DTW) is known to be the best method[ 20 , 21].”