Abstract:
Higher education is a privilege in developing countries like Pakistan where citizens are fighting
even for getting basic education. In the past two decades, more and more students have started
to enroll in IT and engineering related programs in Pakistan but a significant number of these
students dropout before completing their degrees which results into loss of time, money and
seats which could be offered to other deserving students. This problem demands university
administrations and educators to device mechanisms through which student drop out rate can
be controlled, if not totally eliminated. Besides financial support, one such mechanisms which
can help in controlling drop outs is accurate prediction of student performance so that the
students on the verge of failing could be identified and alarmed. This will help them in realising
the efforts needed to show good academic performance. At present, the prediction methods
use academic or cognitive records of students to predict their future performance. Although
the non-cognitive and behavioral aspects are critical in improving student performance, their
role in prediction is yet to be explored. In this research, an effort is made to improve student
performance prediction by predicting performance through combined use of cognitive and noncognitive
features. The result analysis of two different data sets has shown that by adding noncognitive
variables in prediction, prediction accuracies increase using decision tree algorithm;
however the addition does not play significant role in other techniques. The research also
highlighted those individual cognitive features which might help students and educators to
cater for drop outs.
