Abstract
The amount of data in the educational domain is increasing daily with large volumes and complexity which are obtained through the student record systems. Most of these data are under-utilized supporting only simple queries in decision making. whilst there is great interest in their potential use to predict student performance.The aim of this study is to develop a robust model for Predicting student performance. doing the prediction accurately has been a challenging task. it requires the use of machine learning approaches because it was discovered to be successful in various learning tasks. ML consider parameters, and factors that influence student performance. However, predicting student performance with only one classifier is not encouraging because this might be resulting in predictions that are erroneous. Most of state of art studies focus only on the prediction aspect without paying attention to the main reason why a particular model is performing well or not. Hence, majority of students and instructors find it difficult to understand the internal mechanism/structures of how a predictive machine learning model works due to a lack of interpretable and transparent model and this limits the utility of such models. The results obtained from this study show that data modeling and pre-processing procedures are effective in providing a dataset in a suitable format for the analysis and prediction of student performance.
The results also shows that the model parameter tuning is effective in developing a less biased model. In addition, the findings show that feature importance is effective in providing an insight about the role of each variable in predicting student performance. The shap additive method is effective in explaining the internal structure of the model. which makes it easy for relevant stakeholders like students, instructors, and university management in decision making.
The study provides a contribution to knowledge in the student performance prediction domain by investigating variables for predicting student performance and revealed the most relevant and effective variables for the prediction of student performance. The research proposes a conceptual framework for predicting student academic performance and provides a benefit to higher educational institutions in decision-making and policy improvement. An additional contribution made by this study is the development of a reliable and robust model which resolves the issue of biasness from the model parameters, dataset variables, and model interpretability.
As the proposed framework is applied only to the University of Huddersfield, future research should consider expanding the application of this model across other universities. This could assist in ensuring that the generalizability of the model is improved. Comparing single classifiers produced a reasonable result, future research should employ hybrid algorithms for the purpose of investigating whether an improved or similar result could be obtained. the use of a single explainability method, future research should employ more than one method to explain and interpret the model. This could assist in ensuring that the transparency and understandability of the model is diverse and improved. Implementation of this study is somewhat manual, future research should implement a predictive system which is completely automated. This system could assist not only in predicting student academic performance, but also in ensuring that proper support is provided to the students. Since the data variables are academic features, future research should consider features from other categories of student record data This could assist in ensuring that the generalizability of the model is improved.
| Date of Award | 8 Sep 2023 |
|---|---|
| Original language | English |
| Supervisor | Rupert Ward (Main Supervisor) & Richard Hill (Co-Supervisor) |