The huge amount of daily generated data in smart cities has called for more effective data storage, processing, and analysis technologies. A significant part of this data are streaming data (i.e., time series data). Time series similarity or dissimilarity measuring represents an essential and critical task for several data mining and machine learning algorithms. Consequently, a similarity or distance measure that can extract the similarities and differences among the time series in a precise way can highly increase the efficiency of mining and learning processes. We have proposed a novel elastic distance measure to measure how much a time series is dissimilar from another. The proposed measure is based on the Adaptive Simulated Annealing Representation (ASAR) approach and is called the Adaptive Simulated Annealing Representation Based Distance Measure (ASAR-Distance). ASAR-Distance adapts the ASAR approach to include more information about the time series shape by including additional information about the slopes of the local trends. This slope information, together with the magnitude information, is used to calculate the distance by a new definition that combines the Manhattan, Cosine, and Dynamic Time Warping distance measures. The experimental results have shown that the ASAR-Distance is able to overcome the limitations of handling the local time-shifting, reading the local trends information precisely, and the inherited high computational complexity of the traditional elastic distance measures.
Dr. Simon Vilmos
2023-12-29
Támogató: VIK PARIPA program