Predictions of student performance can be useful in many
contexts. For admissions, it is important to be able to identify
excellent students for allocating scholarships and fellowships,
as well as those students who are unlikely to graduate. This
task is extremely difficult with international students, who
come from institutions with diverse grading systems and have
backgrounds that faculty and staff are often unfamiliar with.
The overall prediction accuracy from our analysis was 86%
Session T2G
1-4244-1084-3/07/$25.00 ©2007 IEEE October 10 – 13, 2007, Milwaukee, WI
37th ASEE/IEEE Frontiers in Education Conference
T2G-12
(CTU) and 74% (AIT) for the 3-class prediction. In the case
of admissions decisions for the AIT, the accuracy for
predicting excellent (i.e., B+/A) students was as high as 82%,
and the accuracy for identifying students who would likely fail
was 47%. Therefore, while the system is reliable for
identifying excellent students for the AIT, we will need to
continue to work on the classification problem for identifying
students who are most likely to fail.