The current implementation of the outlier detection is based on the multivariate density. But the 2D projections are so that the outliers may be plot inside the 90% confidence region, because of the projection effect. We should provide an option to detect the outliers based on 2D projections only. This fixes the MatrixPlot from the ProcessHDRAlgo. But it creates inconsistencies, because one trajectory might be outlier on a particular projection, but not on others. The problem is then to plot the outlier trajectories in the physical space.
The current implementation of the outlier detection is based on the multivariate density. But the 2D projections are so that the outliers may be plot inside the 90% confidence region, because of the projection effect. We should provide an option to detect the outliers based on 2D projections only. This fixes the MatrixPlot from the ProcessHDRAlgo. But it creates inconsistencies, because one trajectory might be outlier on a particular projection, but not on others. The problem is then to plot the outlier trajectories in the physical space.