Detecting At-Risk and Withdrawal Students in STEM and Social Science Courses using Predictive and Association Rules Mining

Abstract

This research aims to identify potential at-risk and withdrawal students to help these students in their studies. Interactions consisting of surfing behaviour in the Virtual Learning Environment (VLE) among two different groups of students namely disabled and non-disabled students for Social Science and STEM courses are analysed. Predictive analytics and association rule mining (ARM) analysis are performed. Predictive analytics is performed to predict students’ likelihood of withdrawing from their registered courses. Among the students who choose to pursue their registered courses, predictive analytics is also used to predict at-risk students. Six predictive algorithms namely Decision Tree (DT), Logistic Regression (LR), Naive Bayes (NB), K Nearest Neighbour (KNN), Random Forest (RF), and Support Vector Machine (SVM) are compared. FP-Growth algorithm is applied in ARM analysis. Predictive results show that DT is superior with the accuracy scores reaching 0.91. Most association rules are positively correlated, and they represent the set of commonly surfed pages by the potential at-risk and withdrawal students. The predictive results can help VLE developer to determine the possible algorithms to be used in the intelligent VLE to make accurate predictions based on students’ interactions in the VLE. The results from ARM analysis prove that FP-Growth can also be included in the intelligent VLE. The intelligent VLE can assist the relevant staff in an education institution to provide timely and personalized support to students who are struggling in their studies. This research contributes to precision education through learning analytics.