 
Summary: Monotonicity Hints for Credit Screening
Joseph Silly, Yaser AbuMostafaz
y Computation and Neural Systems program, California Institute of Technology
Pasadena, CA
z Departments of Electrical Engineering and Computer Science, California Institute of Technology
Pasadena, CA
Abstract A hint is any piece of side information about the target function to be learned.
We describe an application of monotonicity hints to a real world problem. The task con
sidered is the screening of credit card applicants. A measure of the monotonicity error of a
candidate function is defined and an objective function for the enforcement of monotonicity
is derived from Bayesian principles. We report experimental results which show that using
monotonicity hints leads to a statistically significant improvement in performance on the
credit screening problem.
1 Introduction
Designing a machine learning model involves balancing two conflicting concerns [1]. The model used
should be powerful enough to implement the details of the unknown target function f , but simple enough
that the optimal parameters for doing so may be estimated accurately from the available data.
The use of hints [2] in the learning process allows us to satisfy the second requirement better without
compromising the first. A hint is any piece of information known about f beyond the available input
output examples. For instance, f may be known to be invariant or symmetric with respect to some
