
236 ◾ Dae-Ki Kang
Ten-fold cross-validation experiments were performed to evaluate human
expert-supplied AVTs on the AVT evaluation data sets used in the experiments for
the AGARICUS-LEPIOTA and NURSERY data sets.
We also evaluated the robustness of the AVTs generated by AVT learner by
using them to construct classiers from data sets with varying percentages of miss-
ing attribute values. e data sets with dierent percentages of missing values were
generated by uniformly sampling from instances and attributes to introduce the
desired percentages of missing values.
Our results for NBL and AVT-NBL revealed the following:
1. AVTs generated ...