In this recent paper the temporal outlier factor (TOF) is introduced for detecting outliers in an unsupervised learning environment. It is simple to compute and yields an "outlier score" for each state space point in the reconstructed state space. So, in fact, the embedding parameters are free parameters of that method.
The question here would be whether the whole TOF-score (i.e. for each point in state space) shall be minimized by the MCDTS or only the minima within the TOF-time series.
The established local outlier factor (LOF) also works on embedding in the time series context and could also be implemented.
In this recent paper the temporal outlier factor (
TOF) is introduced for detecting outliers in an unsupervised learning environment. It is simple to compute and yields an "outlier score" for each state space point in the reconstructed state space. So, in fact, the embedding parameters are free parameters of that method.The question here would be whether the whole
TOF-score (i.e. for each point in state space) shall be minimized by theMCDTSor only the minima within theTOF-time series.The established local outlier factor (
LOF) also works on embedding in the time series context and could also be implemented.