Local Correlation Integral Approach for Anomaly Detection Using Functional Data

Autor: Jorge R. Sosa Donoso, Miguel Flores, Salvador Naya, Javier Tarrío-Saavedra
Jazyk: angličtina
Rok vydání: 2023
Předmět:
Zdroj: Mathematics, Vol 11, Iss 4, p 815 (2023)
Druh dokumentu: article
ISSN: 11040815
2227-7390
DOI: 10.3390/math11040815
Popis: The present work develops a methodology for the detection of outliers in functional data, taking into account both their shape and magnitude. Specifically, the multivariate method of anomaly detection called Local Correlation Integral (LOCI) has been extended and adapted to be applied to the particular case of functional data, using the calculation of distances in Hilbert spaces. This methodology has been validated with a simulation study and its application to real data. The simulation study has taken into account scenarios with functional data or curves with different degrees of dependence, as is usual in cases of continuously monitored data versus time. The results of the simulation study show that the functional approach of the LOCI method performs well in scenarios with inter-curve dependence, especially when the outliers are due to the magnitude of the curves. These results are supported by applying the present procedure to the meteorological database of the Alternative Energy and Environment Group in Ecuador, specifically to the humidity curves, presenting better performance than other competitive methods.
Databáze: Directory of Open Access Journals
Externí odkaz:
Nepřihlášeným uživatelům se plný text nezobrazuje