Mining time-series data using discriminative subsequences

Time-series data is abundant, and must be analysed to extract usable knowledge.Local-shape-based methods offer improved performance for many problems, and acomprehensible method of understanding both data and models.For time-series classification, we transform the data into a local-shape space usinga shapelet transform. A shapelet is a time-series subsequence that is discriminativeof the class of the original series. We use a heterogeneous ensemble classifier onthe transformed data. The accuracy of our method is significantly better than thetime-series classification benchmark (1-nearest-neighbour with dynamic time-warpingdistance), and significantly better than the previous best shapelet-based classifiers.We use two methods to increase interpretability: first, we cluster the shapelets usinga novel, parameterless clustering method based on Minimum Description Length,reducing dimensionality and removing duplicate shapelets. Second, we transformthe shapelet data into binary data reflecting the presence or absence of particularshapelets, a representation that is straightforward to interpret and understand.We supplement the ensemble classifier with partial classification. We generaterule sets on the binary-shapelet data, improving performance on certain classes, andrevealing the relationship between the shapelets and the class label. To aid inter-pretability, we use a novel algorithm, BruteSuppression, that can substantially re-duce the size of a rule set without negatively affecting performance, leading to a morecompact, comprehensible model.Finally, we propose three novel algorithms for unsupervised mining of approxi-mately repeated patterns in time-series data, testing their performance in terms ofspeed and accuracy on synthetic data, and on a real-world electricity-consumptiondevice-disambiguation problem. We show that individual devices can be found auto-matically and in an unsupervised manner using a local-shape-based approach.