Bayesian Network for Predicting Student Retention Based on Expert Elicitation

This project aims to create a Bayesian network (BN) for predicting student retention based on expert elicitation. It aims to construct a Bayesian network (BN) that can be used to predict student retention among and between the variables in the domain.

The model is created primarily by eliciting the necessary information on the types and strength of relationships in the system from the student retention experts.

The participants involved in the elicitation session are experts in their field. It is envisaged that sessions involved interviews and interactive modelling. Session involved a group of domain experts.

The information required from the experts is a causal structure, detailing cause and effect relationships at work in the system being modelled and estimates of probabilities, to capture the strength and characteristic of these relationships.

The values elicited from the experts were used to create the Bayesian network (BN) model. The model was validated by the domain experts.

Academic progress, academic engagement, academic skills, attendance, and financial and moral support from: government, university, partner, family and friends are identified as key variables related to student retention.

Predicting student retention in order to detect students at risk in the early stages of education is essential so that higher education institutions (HEIs) can minimise students not graduating on time or drop out outrightly.

The study shows that in a situation where domain data is sparse or not available at all, the knowledge-driven approach is suitable to be used to elicit estimates of probabilities from domain experts.