Classification of Time-Series Data using ptSTL


Ergurtuna M., Gol E.

28th Signal Processing and Communications Applications Conference (SIU), ELECTR NETWORK, 5 - 07 October 2020 identifier

  • Publication Type: Conference Paper / Full Text
  • Country: ELECTR NETWORK
  • Keywords: Formula Synthesis, STL, Formal Methods, Monotonicity

Abstract

In this work, the goal is to find properties to classify time series data. These properties are expressed using past time Signal Temporal Logic (ptSTL). First, we extend monotonicity properties for signals with timed labels to signals with a single label with the purpose of optimizing parameters of template ptSTL formulas efficiently. This method optimizes a monotone criteria while keeping the complementary criteria (i.e. error) under a given bound. Then by iteratively combining optimized formulas, a classifier is generated for the time series data. Lastly, proposed method is illustrated on a case study.