B0327
Title: Pattern-based transformation for time series classification and anomaly detection
Authors: Marcell Tamas Kurbucz - Wigner Research Centre for Physics | Corvinus University of Budapest (Hungary) [presenting]
Antal Jakovac - Wigner Research Centre for Physics (Hungary)
Abstract: The purpose is to introduce a novel algorithm called pattern-based transformation (PBT) that facilitates uni- and multivariate time series classification and anomaly detection tasks. PBT builds upon the linear law-based transformation (LLT) and utilizes time-delay embedding and spectral decomposition techniques to transform the original feature space. However, unlike LLT, PBT focuses on capturing short-term patterns within the input sequences, resulting in a typically more effective transformation and enabling a more versatile and flexible application of the algorithm. The application of PBT is demonstrated using a wide range of synthetic and real-life datasets, showcasing its effectiveness in various scenarios.
