Um algoritmo baseado em técnicas de agrupamento para detecção de anomalias em séries temporais utilizando a distância de mahalanobis

Abstract

This work proposes an algorithm for anomaly detection in time series data, based on clustering techniques, using the Mahalanobis distance function. After a brief review of the main and recent contributions made in this research field, a formal and detailed description of the algorithm is presented, followed by a discussion on how to set its parameters. In order to evaluate its effectiveness, it was applied to real cases, and its results were compared with another techniques proposed to the same problem. The obtained results suggest that this proposal can be successfully applied to detect anomaly in time series.