Invariant Time-Series Classification

Abstract

Abstract. Time-series classification is a field of machine learning that has attracted considerable focus during the recent decades. The large number of time-series application areas ranges from medical diagnosis up to financial econometrics. Support Vector Machines (SVMs) are reported to perform non-optimally in the domain of time series, because they suffer detecting similarities in the lack of abundant training instances. In this study we present a novel time-series transformation method which significantly improves the performance of SVMs. Our novel transformation method is used to enlarge the training set through creating new transformed instances from the support vector instances. The new transformed instances encapsulate the necessary intra-class variations required to redefine the maximum margin decision boundary. The proposed transformation method utilizes the variance distributions from the intra-class warping maps to build transformation fields, which are applied to series instances using the Moving Least Squares algorithm. Extensive experimentations on 35 time series datasets demonstrate the superiority of the proposed method compared to both the Dynamic Time Warping version of the Nearest Neighbor and the SVMs classifiers, outperforming them in the majority of the experiments.

Long memory and changepoint models: a spectral classification procedure

“…In the context here, it can measure which model best describes a time series, without assuming that this model is where the data were originally generated from. Classification of time series has been previously used in Grabocka et al (2012) and Krzemieniewska et al (2014). It was shown in Yau and Davis (2012) that the autocorrelation function and periodogram of data generated from a changepoint model and a long memory model exhibit similar structures (i.e.…”

Binary Shapelet Transform for Multiclass Time Series Classification

“…The downside to using shapelets is the time complexity. The heuristic techniques described in recent research [15,6] offer potential speed up (often at the cost of extra memory) but are essentially different algorithms that are only really analogous to shapelets described in the original research [17]. Our interest is in optimizing the original shapelet finding algorithm within the context of the shapelet transform.…”

“…The approximate techniques include reducing the dimensionality of the candidates and using a hash table to filter [15], searching the space of shapelet values (rather than taking the values from the train set series) [6] and randomly sampling the candidate shapelets [4]. Our focus is on improving the accuracy and speed of the full search.…”

“…However, it still requires an enumerative search throughout the space of possible shapelets and the full search is O(n 2 m 4 ). Improvements for the full search technique were proposed in [17,12] and heuristics techniques to find approximations of the full search were described in [15,6,4].…”

Variable Star Signature Classification using Slotted Symbolic Markov Modeling

“…Based on the outlined data/domain specific challenges, this paper will attempt to develop a feature space extraction methodology that will construct an analysis of stellar variables and characterize the shape of the periodic stellar variable signature. A number of methods have been demonstrated that fit this profile [21,18,19], however many of these methods focus on identifying a specific time series shape sequence in a long(er) continuous time series, and not necessarily on the differentiation between time series sequences. To address these domain specific challenges, the following methodology outline is implemented:…”