Discrete Wavelet Transform and Maximimum Likelihood Estimation for Anomaly Detection in time series (DWT-MLEAD)

Citekey	ThillEtAl2017Time
Source Code	own
Learning type	unsupervised
Input dimensionality	univariate

Notes

• In contrast to the paper, we use window sizes for the coefficients that decrease with the level number. This makes more sense, because otherwise we would have too few items to slide the window over.
• We exclude the highest level coefficients, because they contain only a single entry, where we cannot slide a window of length 2 over.
• We have not implemented monte carlo sampling for the quantile estimation.

Requirements

• numpy
• sklearn
• pywavelets