Dear all,

I feel confused with a problem. Here is the description. Suppose there is a
data receiver. When it receives a data point, it determines which class the
current data point belongs. If a data point doesn't belong to any
pre-identified class, it is considered as an outlier. One possibility is
that some outliers represent as a whole that a new class emerges; however,
these outliers are not necessarily adjacent to each other. Ideally, the
receiver can modify its outlier detection algorithm by incorporating
information about the new class. 

The problem of concept drift is usually characterized by the assumption of
segmentional stationary, which is not appropriate here. So what kind of
problem it is? Clustering in dynamic environment?

Any suggestion would be highly appreciated.

Best,

Jiaqi

Reply via email to