Abstract
-
It is challenging to test data analytics software
because a test oracle might not be available. This study reports
our experience of applying metamorphic testing to Adobe’s data
analytics software that is used for anomaly detection in a set
of time series data. We make use of geometric transformations
to build metamorphic relations and generate simple time series
data as the source test cases. The results of this study show
that metamorphic testing is highly effective for both verification
and validation purposes. An investigation of the issues detected
during metamorphic testing revealed three bugs in the software
under test.