Using Socio-Demographic Information in Predicting Students' Degree Completion based on a Dynamic Model

Abstract

Educational institutions prioritize identifying variables and features that can help understand students' completion in higher education. This research aims at predicting students' completion based on socio-demographic information. Unlike previous work, a dynamic prediction model was proposed here. Real data of 13262 undergraduate students at a public university in Iraq was integrated and cleaned first. This was followed by applying several feature selection techniques to identify relevant socio-demographic features that can affect the prediction of students' completion. This includes using correlation-based feature subset selection (CfsSubsetEval), symmetrical uncertainty with respect to the class (SymmetricalUncertAttributeEval), GainRatioAttributeEval, ReliefFAttributeEval, and CorrelationAttributeEval. Finally, a prediction model was built based on six prediction approaches which are Bagging, DecisionTable, HoeffdingTree, IBK, J48, RandomForest, and RandomTree. Overall, several different features showed a significant effect on students' completion in which societal regression and the type of secondary school had the highest weight. Furthermore, comparing the accuracy of the implemented classification techniques can reveal that the Bagging classifier outperforms other approaches. The accuracy of predicting students' completion based on this method was 87.56%. Drawing on the research outcomes, educational institutions should pay further consideration to the identified features to ensure students' success and degree completion.