We introduce QuantTree Exponentially Weighted Moving Average (QT-EWMA), a novel change-detection algorithm for multivariate datastreams that can operate in a nonparametric and online manner. QT-EWMA can be configured to yield a target Average Run Length (ARL$_0$), thus controlling the expected time before a false alarm. Control over false alarms has many practical implications and is rarely guaranteed by online change-detection algorithms that can monitor multivariate datastreams whose distribution is unknown. Our experiments, performed on synthetic and real-world datasets, demonstrate that QT-EWMA controls the ARL$_0$ and the false alarm rate better than state-of-the-art methods operating in similar conditions, achieving comparable detection delays.

Change detection in multivariate datastreams controlling false alarms

Luca Frittoli;Diego Carrera;Giacomo Boracchi
2021

Abstract

We introduce QuantTree Exponentially Weighted Moving Average (QT-EWMA), a novel change-detection algorithm for multivariate datastreams that can operate in a nonparametric and online manner. QT-EWMA can be configured to yield a target Average Run Length (ARL$_0$), thus controlling the expected time before a false alarm. Control over false alarms has many practical implications and is rarely guaranteed by online change-detection algorithms that can monitor multivariate datastreams whose distribution is unknown. Our experiments, performed on synthetic and real-world datasets, demonstrate that QT-EWMA controls the ARL$_0$ and the false alarm rate better than state-of-the-art methods operating in similar conditions, achieving comparable detection delays.
Machine Learning and Knowledge Discovery in Databases. Research Track
online change detection, nonparametric monitoring, multivariate datastreams, histograms, false alarms
File in questo prodotto:
File Dimensione Formato  
sub_636.pdf

accesso aperto

Descrizione: Articolo principale
: Pre-Print (o Pre-Refereeing)
Dimensione 351.14 kB
Formato Adobe PDF
351.14 kB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11311/1185541
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? 0
social impact