0
$\begingroup$

I am building a classifier where I have data from two known classes, but I want to capture any new or "unknown" classes as well in my prediction. However, in my case, I do not have any examples of the unknown classes as the data is a multidimensional time-series coming out of sensors, and it's difficult to create such unknown data.

What is the best way to handle such cases?

Improve this question
$\endgroup$
2
  • $\begingroup$ If you formulate "borders" or "radiuses" of the classes in some sense the idea becomes straightforward. This is also the question of "detect outliers/noise" field. $\endgroup$
    – ttnphns
    Commented Jun 6, 2019 at 18:43
  • $\begingroup$ @ttnphns: Can you give any example or technique to formulate such radiuses? $\endgroup$
    – Ash
    Commented Jun 18, 2019 at 17:45

0

Reset to default

Browse other questions tagged or ask your own question.