An innovative big data analytics method for decision makers in the higher education sector to address student attrition

The issue of student attrition in tertiary education has been a significant aspect to develop the core mission and ensure the financial well being of higher education (HE) institutions. The availability of student interaction data via the Learning Management System (LMS) in a blended learning (BL) environment can be utilised to help address the student attrition issue. This thesis presents a novel Big Data Analytics (BDA) method for proving a new provision of utilizing the LMS generated student dataset for transforming new insights related to the identification of struggling students so that decision makers obtain remedial measures that may be taken to mitigate the attrition rate. This study uses a hybrid of Design Science Research (DSR), and Design-Based Research (DBR) approaches to create an innovative integrated DSR research methodology. This methodology is used to develop and evaluate a BDA method. The integrated DSR research methodology prepared the groundwork for the BDA method as the design artefact. The BDA method is founded on a Machine Learning (ML) predictive model trained by utilizing effective ML on a publicly available dataset of student LMS interactions. Identifying students at risk helps to take timely intervention in the learning process to improve student academic progress to increase the retention rate. In the proposed artefact, a variety of ML classifiers are utilised to train and evaluate the predictive model on the original dataset with and without the application of ensemble techniques. Various strategies are explored to further improve the predictive model. Data augmentation and balancing techniques are also applied to the original dataset to balance the sample of each class, and DL algorithms are applied to train and evaluate the models. The BDA method is evaluated to demonstrate its efficiency, accuracy, and effectiveness by a developed evaluation approach. The results show that ensemble techniques, modifying the class distribution, and data augmentation effectively improve the prediction accuracy of the BDA method. The study compares the performance of ML tree based classifiers (Random Forest, J48, NB Tree, Decision Stump, and OneR) and DL algorithms (Multilayer perceptron (ML), Long Short-term Memory (LSTM), and the Sequential Model (SM) neural network). Their performance is evaluated in terms of classification accuracy, specificity, and recall. These results demonstrate the usefulness and effectiveness of ML and DL techniques to construct the BDA method for the identification of at-risk students. The ML based BDA method can be applied to enhance student learning with timely interventions leading to better decision making to reduced attrition rate. The aim of the ML based BDA method is not to replace any existing practices but to support educators to implement various pedagogical practices to improve student learning.