Hubness-Aware Classification, Instance Selection and Feature Construction: Survey and Extensions to Time-Series

Nenad Tomasev, Krisztian Buza, Kristóf Marussy, Piroska B. Kis

Abstract: Time-series classification is the common denominator in many real-world pattern recognition tasks. In the last decade, the simple nearest neighbor classifier, in combination with dynamic time warping (DTW) as distance measure, has been shown to achieve surprisingly good overall results on time-series classification problems. On the other hand, the presence of hubs, i.e., instances that are similar to exceptionally large number of other instances, has been shown to be one of the crucial properties of time-series data sets. To achieve high performance, the presence of hubs should be taken into account for machine learning tasks related to time-series. In this chapter, we survey hubness-aware classification methods and instance selection, and we propose to use selected instances for feature construction. We provide detailed description of the algorithms using uniform terminology and notations. Many of the surveyed approaches were originally introduced for vector classification, and their application to time-series data is novel, therefore, we provide experimental results on large number of publicly available real-world time-series data sets.