Analytics and Information Management in Higher Education

Educational data mining, or the ability to exploit educational data to detect patterns, is an area of increased activity. In this research, we look at the practical application of predictive models in a Higher educational setting in the UK. Firstly, we investigate the use of data mining techniques to highlight performance issues early on and propose remedial actions. We predict good honours outcomes based on data at admission, and some early results from the 1st year. Secondly, we study more granular predictions at the module level. We compare several data mining techniques in order to build both regression and classification models. One of the difficulties we encounter is that, within our problem, missing data is abundant because students do not always take the same module choices. The problem of missing data is prevalent in many data mining applications and remains challenging. We address this problem in a novel way by using multiple imputation combined with an ensemble setting to produce our models. The results show that all the data mining algorithms that use multiple imputation perform better than those without multiple imputation, both in the cases of classification and regression. The algorithms developed, and in particular Support Vector Machines and Random Forest, give us reasonably accurate predictions that could be used as the basis for a future recommender system to assist with module choice selection. Lastly, we study how to use the knowledge found in a way acceptable to students and other stakeholders. For this we design a survey questionnaire to understand student views. We also carry out several interviews with students and some key stakeholders to understand any barriers to change and also to identify enablers. We then analyse the collected data and propose recommendations for the final system