A machine learning-based methodology to predict learners’ dropout, success or failure in MOOCs

• Date: 02 December 2019
• Published date: 02 December 2019
• Pages: 489-509
• DOI: https://doi.org/10.1108/IJWIS-11-2018-0080
• Author: Youssef Mourdi,Mohamed Sadgal,Hamada El Kabtane,Wafaa Berrada Fathi
• Subject Matter: Information & knowledge management,Information & communications technology,Information systems,Library & information science,Information behaviour & retrieval,Internet

A machine learning-based

methodology to predict learners’

dropout, success or failure

in MOOCs

Youssef Mourdi,Mohamed Sadgal,Hamada El Kabtane and

Wafaa Berrada Fathi

Department of Computer Science, Faculté des Sciences Semlalia Marrakech,

Cadi Ayyad University, Marrakech, Morocco

Abstract

Purpose –Even if MOOCs (massive open online courses)are becoming a trend in distance learning, they

suffer from a very high rate of learners’dropout, and as a result, on average, only 10 per cent of enrolled

learners manage to obtaintheir certificates of achievement. This paper aims to give tutors a clearervision for

an effectiveand personalized intervention as a solution to “retain”each typeof learner at risk of dropping out.

Design/methodology/approach –This paper presents a methodology to provide predictions on

learners’behaviors. This work,which uses a Stanford data set, was divided into several phases, namely, a

data extraction,an exploratory study and then a multivariate analysis to reduce dimensionalityand to extract

the most relevant features.The second step was the comparison between five machine learningalgorithms.

Finally, the authors used the principle of association rules to extract similarities between the behaviors of

learnerswho dropped out from the MOOC.

Findings –The results of this work have giventhat deep learning ensures the best predictions in terms of

accuracy, which is an averageof 95.8 per cent, and is comparable to other measures such as precision, AUC,

Recall and F1 score.

Originality/value –Many research studieshave tried to tackle the MOOC dropout problem by proposing

different dropout predictivemodels. In the same context, comes the present proposal with which the authors

have tried to predictnot only learners at a risk of dropping out of the MOOCs butalso those who will succeed

or fail.

Keywords MOOC, Association rules, Machine learning

Paper type Research paper

1. Introduction

The massive open online courses(MOOCs ) are a new model that has revolutionizedthe field

of distance learning since its foundation in 2008 by George Siemens (El-Hmoudova, 2014).

The principle is very simple: open the trainingto anyone regardless of their academic level

or their prerequisites (Sanchez-Gordon and Luján-Mora, 2016). Currently, universities are

increasingly integrating MOOCs, to not only provide training to geographically distant

This research was done through Stanford University’s Center for Advanced Research through Online

Learning (CAROL); the authors are thankful for all the facilities provided. They also wish to express

their full gratitude to Kathy Mirzaei for her responsiveness and collaboration. The authors wish to

warmly thank Mitchell Stevens, Director of Digital Research and Planning, as well as all in the

CAROL Commission for the trust shown to the authors.

Machine

learning-based

methodology

489

Received23 November 2018

Revised7 January 2019

Accepted28 January 2019

InternationalJournal of Web

InformationSystems

Vol.15 No. 5, 2019

pp. 489-509

© Emerald Publishing Limited

1744-0084

DOI 10.1108/IJWIS-11-2018-0080

The current issue and full text archive of this journal is available on Emerald Insight at:

www.emeraldinsight.com/1744-0084.htm

learners but also address issues relatedto the exponential mass of students, as is the case in

developing countries such as Morocco, Tunisia or other countries. In this context, the

MOOCs can also be a tool to complete the initial training, part of which is provided face to

face (hybrid education). MOOCs are not confined to only universities, training companies

offer, in the form of certified MOOCs, freeor paid courses such as Udemy, Coursera, Udacity

and many others (Yuan and Powell, 2013).

Despite their successes,the growth in the number of courses offered and the efforts made

to develop dedicated platforms, the MOOCs have, however,a very appropriate problem that

cannot be neglected: the huge number of learners’dropout (Onah et al.,2014). According to

Liyanagunawardena et al. (2014), only 10 per cent of enrolled learners are able to complete

courses, and therefore, MOOCs far exceed traditional online courses in terms of dropout

rates. For example, a software engineering course offeredby MIT and Berkeley University

received 50,000 registrations but just 7 per cent completed the course (Yuan and Powell,

2013). Another feedbackis from the Open Learning Design Studio of The OpenUniversity of

UK (Cross, 2013), in which the authors studied a coursein which 2,420 student enrolled and

only 50 per cent consulted at least one page of the course during the first week. They also

noted that no more than 30 participantswere active learners, and only 22 people completed

the course, 50 per cent of whom wereable to achieve the course’s objective. Onah et al. (2014)

cited the experience of Duke Universitythat launched a Bioelectricity MOOC, a course that

received 12,175 registrations. Despite this huge number of registrations, only 7,761 learners

(representing 64 per cent of all learners) followedat least one video, 26 per cent answered a

quiz and only 2.6 per cent completed the course.

With the aim to cope with this scourge,a large number of research studies and surveys

have been launched with the ultimate goal to identify the reasons and causes that drive

learners to quit MOOCs. Barak et al. (2016)concluded that language and free MOOCs can be

important factors that drivelearners to drop out. On the other hand, Hone and El Said (2016)

found that the lack of self-motivation among learners and the low quality of learner/

instructor interaction may also cause learners to not complete their courses. Kizilcec et al.

(2017) indicated that the lack in supportand orientation of learners in MOOCs causes a very

high dropout rate.

MOOC platforms generate ample data whose analysis can return relevant indicators

about students dropping out and therefore predictive barometers. Being an important

research topic, several researchers have developed a variety of predictive models by

adopting supervised,unsupervised and semi-supervised machine learning architecturesand

algorithms (Feiand Yeung, 2015).

Predicting learners at risk of quitting the MOOC is an important task, which will allow

MOOC leaders and tutors to present a personalizedintervention to this category of learners.

In this context, it is estimated that in a MOOC, we can find other types of learners, namely,

those at risk of failing and those who will succeed,categories that have not been considered

by the majority of researchers in their proposals or just taken into account separately

(Moreno-Marcos et al., 2018). Classifying learners into three categories will provide tutors

with a more refined view and therefore a rational and effectiveintervention to enable these

learners to follow the MOOC easily. By taking into consideration these three categories of

learners, this paper presents the results of a research that was done on the data set of a

MOOC proposed by Stanford University that had 3,585 enrolledlearners. This research was

divided into four major phases, the firstincluding extraction, preparation and explorationof

data; the second phase was looking for the most predictivefeatures to use when generating

the predictive model; the third phase concerns a comparison of five machine learning

IJWIS

15,5

490