Visual Association Analytics Approach to Predictive Modelling of Students’ Academic Performance

Published Online December 2019 in MECS DOI: 10.5815/ijmecs.2019.12.01

The increasing dependency on technology and methods that are driven by information technology by academic institutions has accounted for the abundance of huge educational data repositories. Moreso, educators and educational administrators have intensified their efforts towards collecting and storing data providing information on the functionality of their educational systems. These repositories have the capacity of storing a large amount of student-related data and information. Students’ data and information are key requirements in educational learning systems, especially in the planning, monitoring, and assessment of educational management systems (EMS). EMS is a platform for data acquisition and collection, validation and processing, analysis and communication of information relating to administrators, students, teachers, staff and infrastructure in the educational environment [1]. EMS is a rapidly progressing field of data mining which concentrates on the search and discovery of interestingly new patterns, techniques, tools, and models for intelligent exploratory analysis and visualization of large educational dataset. EMS is aimed at the extraction of novel and interpretable structures that will enhance comprehensibility of students, their processes and environments [2,3]. Among the important modules of EMS are students’ management (SM),  human resource, infrastructure management, school  management, and graduands’  management module. The SM module captures and stores students’ data with a unique student enrolment number as the primary key, demographic data, academic status amongst others. Associated with EMS, is learning analytics (LA) and educational data mining (EDM), concerned with the exploratory analysis of the educational dataset and utilizing the outcome directly on the students, teachers and other components in the learning process. EDM and LA try to interpret how students cope and interact with educational resources at their disposal, their learning behavioural patterns, likely final academic outcome, and most importantly, their likelihood of completing their program within the minimum stipulated timeframe. EDM depends more on methods, tools, and techniques while LA focuses on the description of data, knowledge and resultant patterns. Machine learning techniques, statistics, visual analytics, link analysis, opinion, and sentiment analysis are some of the widely used tools of LA while classification, clustering, Bayesian modeling, relationship mining, discovery with models and predictive/prescriptive modeling are often associated with EDM [4]. Predictive analytics employs a range of statistical approaches ranging from machine learning and predictive modelling to data mining to competently analyze the historical and operational data and information to enable predictions about the unidentified future event. Its application cut across several domains including academic performance prediction.

Academic achievement is a key factor considered by recruiting organizations and motivates the monitoring of students’ performance during their academic pursuits. Students have to work hard for outstanding grades, in order to rise up to the potentials of recruiting organizations and meet the expectations of parents/guardians, educators and administrators. Persistent and quality graduation rates of students are increasingly important indicators of progressive and effective educational institutions. Any educational system characterized by high rates (frequency) of drop-outs (students who leave an institution precipitately without completing the desired programme of study), transfer-outs (students who started in one course of study or one institution and, thereafter move to another course or educational institution to enable him/her graduate) stop-outs (students who voluntarily withdraw and leave for a period of time, and then re-enroll in order to complete their programme), and spill-over (students who spend extra year(s) in due to poor performance) is said to fail [5,6]. The early identification of students’ weaknesses during their academic career will guide in the effective provision of necessary pedagogical interventions, suggesting behavioural changes to enhance students’ learning processes and also ensure students’ on-time and satisfactory graduation [7-9]. However, educational systems in most developing countries) lack facilities for automatic predictions of fail or pass percentages of students and cannot account for the number of drop-outs, stop-outs, transfer-out or spill-over students but rather concentrate more on successful students. They have no information about what patterns lead to these at-risk students and cannot identify students who are likely to struggle in their academics at an early stage of their academic pursuits. In consideration of these challenges, this paper employs EDM and LA methodologies (cluster analysis and association rule mining) to model students’ learning processes, for informed decisions and timely pedagogical interventions. Association rule mining [1]. [10] attempts to extract relevant and interesting relationships among items in a database. This paper aims at identifying relationships among courses offered by the students, and the effects of such correlations to learning and academic performance vis-a-vis status. It will also employ cluster analysis to reveal the optimal number of course clusters and their association with student’s status at the end of the minimum duration of the programme.

The rest of the paper is organized as follows. Section II presents literature review with emphasis on cluster analysis and performance-course association rule mining. In section III, the methodological framework is conceptualized for the implementation of the system. Course association rule mining procedure and results are described in Section IV. Discussion of results, and conclusions and further work is presented in section V and VI respectively.

• II.    Literature Review

Learning analytics and EDM can discover and extract trends in data, and also act as a medium for promoting educational activities by identifying and avoiding failure (or poor performance) trends and patterns while exploiting and utilizing success patterns. In Ref. [11], EDM and learning analytics promise to make sustainable impact on learning and teaching to transform slow learners into effective and better learners [12]. Reference [13] points out that learning analytics involves two major operations namely predicting student learning successes and providing proactive feedbacks. Reference [14] proposed a multivariate based method of predicting students’ results in learning courses associated with web learning while reference [15] reported that, to make sense of large amounts of educational data, intelligent systems must be developed to automatically process the data and provide reports to stakeholders. In reference [16] a LA dashboard to enhance students’ learning performance was developed. The system in reference [16] works by tracking and mining massive online student data and visualizing results so they can be comprehended at a glance. Experimental evaluation indicates that although the LA model did not have a significant impact on student achievement, there was an overall student satisfaction with the dashboard which impacts on students’ understanding level.