(1144) MINING ENROLMENT DATA FOR EARLY IDENTIFICATION OF ATRISK STUDENTS

摘要

This study uses predictive modelling, one of the data mining techniques, to help identify at-riskstudents (which is defined as those whose cumulative grade-point average (CGPA) of <=2.3 out of amaximum attainable value of 5.0) in their first semester in SIM University (UniSIM), which catersmainly to part-time adult learners. The objective of identifying these students early is to enableintervention to be effected as soon as the students are enrolled and this could be before they evenembark on their studies in their first semester.11 variables from the enrolment database were considered as possible factors to the predictive model,which can be divided broadly into demographic variables (eg. age, gender, years of workingexperience), pre-UniSIM variables (e.g. polytechnic graduated from, polytechnic CGPA, 'O' level Mathand English grades) and UniSIM variables (e.g. School/Discipline enrolled in UniSIM). To classify theat-risk students, various algorithms were used e.g. neural network, and decision tree. Theperformances of the various models were compared using sensitivity, specificity and accuracy and thechosen model is a decision tree model that was also able to inform on policy. The chosen decisiontree model identified the following factors: polytechnic that the student graduated from, polytechnicCGPA, O' level Math and English grades, the School that the student is enrolled in and the year sincethe student graduated from polytechnic. The implications of these results for identification of earlyintervention are discussed.