A possible interesting test which was not performed is to test the systems performance on data sets with missing attribute values. This could follow the test patterns above, and check the classification accuracy when attributes are missing both in the test and training sets. Multiple classification methods should be used with various numbers of missing attributes.