Outlier Detection Based on Wavelet-HMM methods
Download as PDF
DOI: 10.25236/icidel.2017.079
Author(s)
Fang Liu, Ruzhen Dou, Chaoying Xia
Corresponding Author
Fang Liu
Abstract
According to the limitation of the principle of outlier detection based on wavelet, this paper proposes a new outlier detection method called Wavelet-Hidden Markov Model (W-HMM) algorithm. In this algorithm, the signal is decomposed in some scale, and when the wavelet decompositions of the signal are different from the most other wavelet decompositions, the signal will be seen as potential outlier. Aiming to make further accurate judgement, the similarity measure between the wavelet coefficient of this signal and that of normal signal will be done, and the final confirming is made by Viterbi algorithm which is used to HMM. The validity and practicality are proved by experimentation and application in this paper.
Keywords
Outliers detection, Improved recursive wavelet transform, HMM, Process data.