Data mining techniques for identifying students at risk of failing a computer proficiency test required for graduation

Abstract

<blockquote>Enabling undergraduate students to develop basic computing skills is an important issue in higher education. As a result, some universities have developed computer proficiency tests, which aim to assess students' computer literacy. Generally, students are required to pass such tests in order to prove that they have a certain level of computer literacy for successful graduation. This paper applies data mining techniques to make predictions about students who are going to take the computer proficiency test and fail. A national university in Taiwan is considered as the case study. Three different clustering techniques are used individually to cluster students into different groups, which are k-means, self-organising maps (SOM), and two-step clustering (i.e. BIRCH). After the best clustering result is found, the decision tree algorithm is used to extract useful rules from each of the identified clusters. These rules can be used to warn or counsel students who have higher probability of failing the test. The results can help the university identify a number of student groups who need to pay much more attention to preparing for the test, which is likely to help conserve resources. Furthermore, this study can be regarded as a guideline for future developments in assessing students' English literacy, as this is also an important graduation requirement in many universities.</blockquote>

The efficacy of learning analytics interventions in higher education: A systematic review

“…That is, these types of models incorporated, for instance, high school GPA, socio-economic status (SES), and scholastic aptitude test (SAT) scores in an algorithm to forecast student success or retention at a designated future time -such as at the end of a course in the first year of university (e.g. Agnihotri & Ott, 2014;Cochran, Campbell, Baker, & Leeds, 2013;Dekker, Pechenizkiy, & Vleeschouwers, 2009;Green, Plant, & Chan, 2016;Guruler, Istanbullu, & Karahasan, 2010;Harrak, Bouchet, Luengo, & Gillois, 2018;Kotsiantis, Pierrakeas, & Pintelas, 2003;Morris, Wu, & Finnegan, 2007;Tsai, Tsai, Hung, & Hwang, 2011;Yasmin, 2013;Yukselturk, Ozekes, & Türel, 2014). Traditional LA models thus included a relatively simple combination of student characteristics at one time-point to predict later academic performance (Williams, 2014).…”

Examining Techniques to Solving Imbalanced Datasets in Educational Data Mining Systems

“…The study concluded that students' culture or ethnicity is one of the main factors affecting students' performance. Tsai et al [17] applied the K-means algorithm, unsupervised neural networks, and the C5.0 decision-tree algorithm at the National University in Taiwan to cluster and predict the undergraduate students. The purpose of the study was to develop an earlywarning system to identify the students who might fail one of the graduation requirement tests, the computer proficiency.…”

Exploring the Colombian digital divide using Moodle logs through supervised learning

“…For example, Xu and Yang (2016) use data involving learner activities such as video play activity and course forum activity from MOOCs on Coursera to predict student motivation and whether they will obtain a certification, using SVMs. Tsai et al (2011) evaluate the ability of various clustering techniques to derive decision trees to assess students’ computer literacy. Lopez et al (2012) analyse student participation in Moodle forums, including variables such as number of messages sent, number of messages read on the forum, time spent on the forum and degree centrality of the student, to estimate the final mark.…”