Abstract
We present the system nFOIL. It tightly integrates the naĻive
Bayes learning scheme with the inductive logic programming
rule-learner FOIL. In contrast to previous combinations,
which have employed naĻive Bayes only for post-processing
the rule sets, nFOIL employs the naĻive Bayes criterion to
directly guide its search. Experimental evidence shows that
nFOIL performs better than both its base line algorithm FOIL
or the post-processing approach, and is at the same time competitive
with more sophisticated approaches.

(Q1) Is there a gain in predictive accuracy of nFOIL over
its baseline, FOIL?
(Q2) If so, is the gain of an integrated approach (such as
nFOIL) over its baseline larger than the gain of propositionalization
approaches?
(Q3) Relational naĻive Bayes approaches such as 1BC2 essentially
follow a propositionalization approach, employing
all features within the bias. Does nFOIL employ less
features and perform well compared to these approaches?
(Q4) Is nFOIL competitive with advanced ILP approaches?


http://www.informatik.uni-freiburg.de/~ml/papers/aaailandwehr2005.pdf