It is very easy! You just have to create a new dataset excluding the anomalous instances.
If you are using the BigML Dashboard, in the anomaly view, you need to first select the anomalous instances you want to remove (all top ten anomalies in the example below) and then click the highlighted icons before you click the Create dataset button:
If you are using the BigML API you will have to exclude the anomalous instances by filtering your rows as explained in this subsection of the documentation for developers. You can write your filter using JSON syntax or Lisp syntax and set it as "json_filter" or "lisp_filter". Please click here for more details.