In a very recent paper it is suggested to obtain an adaptive recurrence threshold based on local maximum attractor densities/distances. Basically, the distance matrix gets scanned using a sliding window (window size is a free parameter, but can be related to the decorrelation time of the time series) and for each window the maximum distance is picked. In turn, this local maximum distance will be used to scale the recurrence threshold accordingly.
The paper and the shown results sound promising and the effort to implement it (and also the computational complexity itself) seems to be manageable I suppose.
In a very recent paper it is suggested to obtain an adaptive recurrence threshold based on local maximum attractor densities/distances. Basically, the distance matrix gets scanned using a sliding window (window size is a free parameter, but can be related to the decorrelation time of the time series) and for each window the maximum distance is picked. In turn, this local maximum distance will be used to scale the recurrence threshold accordingly.
The paper and the shown results sound promising and the effort to implement it (and also the computational complexity itself) seems to be manageable I suppose.