Impact of LMS Selection on Students’ Activity

Students’ Activity Evaluation Problems in Moodle and Open edX Learning

Management Systems

Viktors Zagorskis and Atis Kapenieks

Distance Education Study Centre, Riga Technical University, Kronvalda 1, Riga, Latvia

Keywords:

LMS, MOODLE, Open edX, UX.

Abstract:

The quality of youth education, in general, is worsening (Coppola and O’Higgins, 2015). Some of the reasons

for such a statement are (i) the current changes of educational systems in general, (ii) eventually unreasonable

switching between ephemeral learning trends and tools. Despite the use of learning management systems,

learners or educators, face today a real and increasing difﬁculty in ﬁnding optimal communication for the

learning coalition between the students and the teachers. In this research, students’ activity in two popular

learning management systems (LMS): MOODLE and Open edX are analyzed. We examine both platforms by

measuring and comparing the students’ activity from the learning-time aspect. The novel – CAST algorithm

is proposed to reduce the overestimated learners’ activity measuring errors, caused by unterminated WEB

sessions. We conclude that Open edX engage students more, therefore, educational institutions should move

forward to modern eLearning platforms.

1 INTRODUCTION

Today, in the era of global networking the interaction

between students and instructors is organized mostly

in a blended form - in class and through eLearning

systems (LMS).

Organizational aspects of teaching have effects on

learners. Strategies that require students to be acti-

vely engaged with the learning material produce bet-

ter long-term retention than the passive act of reading

(TULVING and Craik, 2000), (Benjamin, 2011). Si-

multaneously, there is need for some difﬁculties in le-

arning (Bjork, 1994).

The educational institution is responsible for in-

vention and support of learning strategies – inclu-

ding blended learning, class organizational approa-

ches, and platform(s) for e–learning.

At Riga Technical University (RTU), Moodle

LMS (Romero et al., 2008) since the year 2007, is

recommended for application in hundreds of lear-

ning courses. The Distance Education Study Cen-

tre (DESC) is responsible for the supervision of the

subject ”Basic Business”. Every academic year from

2013 to 2017, this course was conducted for more

than one hundred ﬁrst–year students. More than 600

learners in the ﬁve year period were addressed.

Thanks to the DESC staff qualiﬁcation, new hori-

zons for education using modern communication de-

vices together with multi-screen learning technolo-

gies (Kapenieks, 2013) were presented. At the mo-

ment Open edX is the primary open source appli-

cation to support MOOCs (Massively Open Online

Courses) (Pijeira D

ıaz et al., 2016). The commu-

nity using Open edX is growing. During the last two

academic years (2016, 2017) the subject ”Basic Busi-

ness” was reconstructed and is maintained entirely on

Open edX LMS.

Learning, in general, is a stochastic process to-

ward the learning goals. Some theories (Mosteller,

1958) for decades are used to model such proces-

ses. Among the authors, Bartholomew’s work on sto-

chastic processes provides the theoretical foundation

for stochastic models of learning (Ifenthaler and Seel,

2012).

In spite of guided training in blended learning

(Seel, 2012), controlling the students’ diversion of at-

tention from the learning content is still problematic.

Also, impact of random, hidden factors decreases by

attractive and personable learning content (Cornelius-

White and Motschnig, 2012).

Today, in modern massive online LMSs like

UDEMY, UDACITY, COURSERA, EDX, etc. stu-

dents are often engaged using intuitive, user expe-

rience (UX) oriented navigation, short text or video-

Zagorskis, V. and Kapenieks, A.

Impact of LMS Selection on Students’ Activity.

DOI: 10.5220/0006810205050512

In Proceedings of the 10th International Conference on Computer Supported Education (CSEDU 2018), pages 505-512

ISBN: 978-989-758-291-2

505

based learning content (Cennamo, 2012) (Maniar,

2012), gaming based (Seel, 2012) learning contents,

as well as regular following the obtained scores.

Among others, the earned scoring observation is good

motivation factor to learn.

In blended learning, such an approach motivates

students to use eLearning tools more, although it can

lead to mental overload of the learner. Interaction

with an increasing number of networked devices, ap-

plications, and web services; learner attention is the

new bottleneck in computing (Okoshi et al., 2015) and

the learning domain too.

Since the knowledge items in MOODLE and

Open edX are for sure presented technically diffe-

rently (styles, forms, the interface of access, etc.), we

realize that objective comparing of systems is a tough

problem. Although, we are motivated to know more

about user behavior difference in various LMS.

Our goal of this research is to explore the impact

of switching from Moodle to Open edX on students’

activity, keeping the learning content (number of to-

pics, problems, assignments, and assessments) un-

changed. To achieve that, we study and analyze data

retrieved from both LMS by:

1. mapping students’ activity to the learning-time,

2. applying of approximation algorithm that reduces

learning-time overestimation problems,

3. comparing user behavior in both LMS using ex-

ploratory analysis.

The paper is organized as follows: Section 2 intro-

duces the data acquisition principles, tools, formats,

and limitations. Research models, methods, and pro-

posed algorithms are classiﬁed in Section 3. Section

4, which is the core section of this paper, provides an

analysis of results of learner activity computation.

Finally, Section 5 offers the conclusions of the re-

search.

2 DATA GATHERING

We start with decision-making regarding the appro-

priate strategies, tools and methods for learners data

gathering and analysis based on application domain.

2.1 Application Domain

Since 2000, the data mining paradigm has been often

used. As stated (Kumar and Reinartz, 2006), ”data

mining provides businesses with the ability to make

knowledge-driven strategic business decisions”. In

our case, the business-application domain is ”Blended

Learning” (in class/self paced). The business decision

is about the selection of a more efﬁcient e-learning

system from two available for experiment: Moddle

and Open edX. We also take into account the corre-

spondence of tools to automate some steps in the data

science process in the future.

2.2 Data Source Domain

Data, used for students’ activity evaluation was mined

from MOODLE, and Open edX log ﬁles. In our rese-

arch experiment, we used two consecutive hold cour-

ses (years 2015 and 2016). Courses were organized

in the period from 1 of September till 31 of Decem-

ber. LMS allowed access to the contents all day long

(24/7). Therefore, total course contents’ access time

(Course Period) physically was 2929 hours.

Each course had the same amount - 126 learning

objects. On average, course staff expect that students

could spend no more than ﬁve minutes attention to

each learning object observing it only once. So, the

time needed for students just to log into LMS, to stay

logged in, and introduce themselves to all the items is

about (126 items * 5 minutes = 10.5 hours) per whole

course contents. Here, we assume that each student

pay attention to all the objects – but only once. No-

tably that 10.5 hours of learner activity in the LMS

is the minimal expected value and is equal to 0.36 %

from the whole Course Period. In reality, we argue

that each learning object author expects that attention

to his/her creative work will be given more than once.

The number of students in Moodle was 114, whe-

reas 148 in Open edX. Moodle course students pro-

duced 22775 logged activity records versus 70175 in

Open edX (see Table 1).

Table 1: Data amount and producers.

Description Moodle Open edX

Course Period [hours] 2929 2929

Technical Length [hours] 10.5 10.5

Learning Objects 126 126

Number of Students 114 148

Logged Records 22775 70175

Since experimental courses, we are analyzing,

were organized using Blended Learning strategy, the

most time - 32 hours learners spent in the class. Here,

we had no ability to measure individual time used for

individual learning without LMS usage.

2.3 Data Source Model

LMS logged data follows the Time Series data source

concept. In general, Time Series analysis comprises

two methods: (1) to extract meaningful statistics of

CSEDU 2018 - 10th International Conference on Computer Supported Education

506

the data, (2) to forecast future values based on previ-

ously observed values. As modern direction in Time

Series studies is deep neural networks. Although,

such an approach is computational resource consu-

ming.

In this paper we study Time Series data using sta-

tistics based technique: data exploratory analysis and

visualization.

2.4 Data Analysis Tools

After LMS examination, we identify - the learners’

data is stored in ASCII type log ﬁles with access on

demand. Files’ sizes are a couple of ten megabytes in

compressed form. In this case, as a tool for data acqui-

sition, cleansing, presentation, and modelling we use

R programming language (Wickham and Grolemund,

2016).

2.5 Data Formats

In this subsection, we shortly introduce data formats

and their differences between Moodle and Open edX

before approximation algorithm is applied.

Logged Data from the LMS ﬁles are used. The

plain log ﬁles are keeping records, previously conﬁ-

gured by LMS developers that were guided by archi-

tecture documentation.

We handle plain ﬁles that keep learners’ data logs.

The key features of Logged Data File formats for

either LMS are:

• Each record line corresponds to one logged acti-

vity produced by a learner or a staff member or a

system.

• Logged Data records for Open edX are stored in

JSON (JavaScript Object Notation) format ﬁles,

Each logged event is recorded as a new JSON list

object.

• Logged Data, generated by Moodle is structured

by the record position in a text line. Each logged

event is recorded as a new text line.

2.6 Pre-processing Data

Data, generated and recorded by the personnel, or

course staff, or system are to be dropped in the cle-

ansing process. To do that, we apply different data

acquisition techniques to the raw data. Processing

methods, borrowed from R programming language

documentation and best practices (Zumel and Mount,

2014) used to separate records and present them in a

one-dimensional vector form.

The data acquisition process begins similarly in

both LMS. Here, the single raw Logged Data File

record line is scanned, transformed, and represented

as ”moodle data” or ”edx data” into character type

vectors using R function scan.

Each (moodle data or edx data) vector element

consists of various character groups and LMS speciﬁc

separators. Pre-processing routines allows to trans-

form data to data tables suitable for further analysis.

Here, we show the example of a Computer Pro-

gram code in R that scans data and process into table:

moodle_data <- preprocess(scan(moodle_file))

edx_data <- preprocess(scan(edx_file))

As a result of pre-processing, we get clean data in

table format.

2.7 Data Structure

Moodle and Open edX have different data structures.

In our experiment we identiﬁed that Moodle LMS has

only 6 data categories, whereas Open edX has an ex-

tended structure consisting of 10 categories. Some

Open edX categories have ”context” consisting of 7

subcategories. Obviously, Open edX wins by the

amount of data logged.

2.8 Data Precision

Example formats of time stamps for both LMSs are

given in the following code lines:

Moodle: 2017,January,23,07:21

Open edX: 2017-01-23T07:21:17.048129

The signiﬁcant potential of Open edX is revealed

through more precise event time stamp logs in a com-

paring with Moodle. Here, we ﬁnd that Open edX is

better not only in the case functional analysis of lear-

ners’ interaction style with video-based content, but

also is ready for more precise methods’ invention.

2.9 Summary

Comparing of Data structures from both LMS we

conclude:

• Open edX has more distinct log record categories

out of the box: 16 vs. 6 for Moodle,

• the data format used in Open edX is JSON, which

makes further data searching, separation, and ana-

lysis more efﬁcient due to simple coding.

• time events logging precision for Open edX has a

resolution expressed in microseconds.

• logging of interactions with video content has no-

table intelligence, allowing to provide further spe-

cialized user behavior analysis.

Impact of LMS Selection on Students’ Activity

507

3 TOWARDS LEARNER

ACTIVITY MODEL

In this section we describe the data gathering work-

ﬂow along with the developed algorithm that help

us to verify the positive impact of switching from

Moodle to Open edX on learners activity.

Firstly, using raw Logged Data Files from both sy-

stems, we apply some methods to obtain, clean, and

transform learners’ activity information from Time

Series into similarly structured (data frames). Next,

we apply in Section 3.2 proposed CAST algorithm to

pre-processed data. The output of the algorithm ope-

ration is a new data set EDS (Estimated Data Set).

The EDS set is exploratory ready and contains our es-

timations. EDS has data that takes into account user

behaviour uncertainties that in WEB environment can

not be explicitly identiﬁed. More precise, we are mo-

delling user activity interruption without the identify

of WEB session end. So, computed EDS contains a

number of estimated activity paths for each student.

Then, we apply some visual data exploring methods

to obtain scatter plots and histograms from EDS data.

Finally, we evaluate statistical differences between

users’ sessions in Moodle and Open edX.

Although, the trivial learner activity model - using

WEB sessions’ beginning and ending records, seems

to be applicable, this assumption can lead to the seri-

ous model errors. Overestimation is the main reason.

3.1 Overestimation Problem

By observing students working at the computer in

class, we conclude, that after opening the session to

LMS, learners often – after a couple of minutes, turn

their thoughts to other life events. Therefore, stu-

dents, as stated (Okoshi et al., 2015), reduce cumu-

lative, learning content oriented attention due to some

uncontrollable reasons.

By simple approach, students’ online activity time

spent in the LMS can be measured using uncomplica-

ted session time-delta computation (Figure 1).

Figure 1: Log record based session length computation.

If all the time that students are sitting in the front

of the computer we are taking into account, learners

activity becomes overestimated. In addition, we attri-

bute distracted user behavior to learning apart of the

group.

Therefore, we assume that straight reﬂecting of

online web session length, computed by identiﬁcation

of session start-stop logged facts is misleading. The

main argument is learner distraction observed in class,

also mentioned in the previous paragraphs (Okoshi

et al., 2015).

In the following example, the session is split into

Active and Idle parts (Figure 2).

Figure 2: Session reduction due to learner distraction.

The given ﬁgure reﬂects the simplest possible

case. In reality, different idle part patterns can be wit-

nessed: (a) student behavior patterns, and (b) techni-

cal issues, like connection lost, battery downtime, etc.

To avoid students’ activity time overestimation

preparing data for analysis, we introduce a novel al-

gorithm, named CAST.

3.2 Introduction to CAST Algorithm

We propose the CAST algorithm that computes Cu-

mulative Activity and Session Time approximation

for the eLearning environment. In the case of unter-

minated WEB sessions (Sikos, 2011) the CAST algo-

rithm is suitable for estimating the learners’ activity.

The main idea of CAST estimation algorithm is

looping through all students’ records creating indivi-

dual learner activity proﬁle (see Algorithm1).

The CAST algorithm ignores possible ’Session

End’ records. The session length for the investigated

learner is computed using an invented vector with the

fake ’Session End’ values. The algorithm counts the

learner sessions and marks sessions in output table as

terminated based on user activity in the system.

The CAST algorithm has two Input groups of pa-

rameters. The algorithm analyses already cleansed

LMS Log Data, that is the ﬁrst group for the Input

data (see Figure 3). The second data component is

the artiﬁcially created Maximum Session Length vec-

tor (MSL). MSL vector has numeric values can be va-

ried by the algorithm operator to cover the full range

of approximated learner activity model in the LMS.

The Out put of The CAST algorithm is the Estima-

ted Data Set (EDS), consisting of two data frames H

and G. The H data frame is the output of ﬁrst CAST

CSEDU 2018 - 10th International Conference on Computer Supported Education

508

Figure 3: CAST Algorithm Inputs and Output.

algorithm method A. The intermediate result of met-

hod A computes the set of approximated data for each

user individually. The G data frame is produced by

second - sequentially executed method B that avera-

ges data for the whole learners’ group. The G data

frame stores nine data vectors that summaries lear-

ners’ group activity in the LMS through (a) number

of counted activity periods - named sessions, and (b)

cumulative activity time over the whole course period.

Table 2 shows G data frame column names. Here, we

use following notation: CATH stands for ”Cumulative

Activity Time, expressed in Hours”.

Table 2: G data frame column names.

Name Description

MSL maximum session length

CSA

min

min of counted sessions

CSA

max

max of counted sessions

CSA

mean

counted sessions mean

CSA

median

counted sessions median

CAT H

min

min cumulative activity time

CAT H

max

max cumulative activity time

CAT H

mean

cumulative activity time mean

CAT H

median

cumulative activity time median

Either LMS (Moodle or Open edX) has it’s own

EDS data set on output.

The newly computed EDS subsets (H and G)

helps to reveal the most reliable session length by ap-

plying some general statistical analysis methods, dis-

cussed in the next Section 4.

3.3 CAST Algorithm Description

We assume that learners produce the principal amount

of logged events by simply clicking the mouse on

the different computer or mobile device screen area.

LMS stores event logs. CAST algorithm use these

events later in EDS data computation.

Firstly, for CAST operation we create the MSL

vector with Maximum Session Length parameters. As

a trade-off, to safe computation time and get visually

ﬁgured out data, we use steps of 5 minutes. In practice

we use the range between 5 and 40 minutes storing

these values in the R ”vector” MSL:

MSL <- c(5, 10, 15, 20, 25, 30, 35, 40)

Next, we load the chosen LMS Log Data into the R

”dataframe”, namely the CourseData. Now, we de-

ﬁne the course time boundaries: StartDate, EndDate,

and compute the Course Technical Length (CT L) as

the difference between the course start and course end

dates. After computation, we get the course length

equal to 2929 hours.

Now, we identify records for time-series events’

study. Finally, we create two vectors with 114

(Moodle) or 148 (Open edX) unique learners’ (user)

names (see Table 1).

The ﬁrst method A(M, S, E) of CAST algorithm

(Algorithm1) needs three input parameters to operate.

The ﬁrst - M have to be created by the operator of the

algorithm. The S (list with unique students’ names)

and E (logged activity events) are prepared from log-

ged data.

Algorithm 1: CAST algorithm.

Method:

Input: S : the set with unique students’ names

E : the set of logged events’ time-series

M : the MSL vector M = {5, 10, . . . , 40}

Output: H : set of individual approximated data

/* Step 1. compute estimated data for all users*/

1: for all m ∈ M do

2: for all s ∈ S do

3: cat ← 0  cumulative activity time

4: k ← 0  activity periods’ counter

5: τ ← 0  present period length

6: for all e ∈ E do

7: t ← Di f f (E

− E

n−1

)

8: if t < m then

9: τ ← τ + t

10: else

11: τ ← 0

12: k ← k + 1  count the session

13: end if

14: cat ← cat + τ  add to total time

15: end for

16: H[s] ← s[cat][k]  estimated cat and k

17: end for

18: H[s][m] ← m  store data at given MSL

19: end for

20: return H

Method: B(H)

Input: H : the set of individual estimations

Output: G : the summary on group of learners

/* Step 2. compute summary/average estimators */

21: G[s][summary] ← S ummary(H)  like min, max

22: G[s][average] ← Average(H)  mean, median

23: return G

Impact of LMS Selection on Students’ Activity

509

Applying the ﬁrst method A, the CAST algorithm

computes estimated activity data for all users in the

list S by the given M vector value.

The outcome of the A method is a data frame

H[s][m], containing estimated data [s] for each student

in the S list at the given M set value. The [s] data con-

sist of two subsets: cumulative activity time cat, and

counted activity periods k (line: 16).

The key of the algorithm is the inner loop (lines:

6 to 15). By iterate all recorded events, algorithm me-

asures the time difference (inter-event-time) between

logged activities (line: 7). If a current time difference

between nearby events is less than the suggested MSL

value, we assume that a student is still active. We

increase student’s present-active time period by the

value of the time difference measured from the pre-

vious event (line: 9). If a student has not renewed

his activity for more then given MSL value, we count

student’s session as terminated (line:12) and reset pre-

sent activity time, making ready for usage in the next

session.

After the application of all student speciﬁc re-

cords, taken from the logged events, the student’s

approximated activity expressed through cumulative

operation time measured in hours (CATH) and ap-

proximated number of activity periods (or sessions)

is stored in output data set H (line: 16).

Looping through the MSL values let us to store

all the users’ behaviour proﬁles as a function of MSL

(line:18).

By applying of B method to the H data, we get sta-

tistical summary G data set that helps us to compare

Moodle and OpenedX LMS.

4 DISCUSSION OF THE RESULTS

In this section the EDS data is presented through the

visualization of H data subset using statistical evalu-

ations, and as scatter plots, revealing G data subset.

4.1 Sessions’ Count

The plot visually depicts G subset as ”mean” of the

number of counted sessions of the whole course ver-

sus the modeled MSL value (Figure 4). Here, we

assume that the expectation about the errors is sym-

metric. The ”median” of data is also used later as a

consistent estimator and is less sensitive to extreme

observations (DeGroot and Schervish, 2013).

From Figure 4, we can identify that Open edX is

used more. The numeric difference is computed from

G data subset and is (in average) per ﬁve sessions

more in the comparison to Moodle LMS.

We use median estimator in Figure 5 for explo-

ring the experiment data under the assumption of the

existence of extreme data observations. Here, we in-

tuitively ﬁnd by the median estimator depicted time

boundary of Moodle usage - no more than 20 minutes

on average.

0 10 20 30 40 50 60

15 20 25 30

Maximal Session Length (MSL) [min]

Counted Sessions (by CAST)

MEAN values:

edX

MOODLE

Figure 4: Mean value of counted sessions (from subset G)

as a function of a modeled MSL value.

0 10 20 30 40 50 60

10 15 20

Maximal Session Length (MSL) [min]

Sessions Counted by CAST

MEDIAN values:

edX

MOODLE

Figure 5: Median value of counted sessions (from subset G)

as a function of a modeled MSL value.

4.2 Comparing the Means

We compare two independent data groups: Moodle

and Open edX. For comparison of means of G data

subset (Table 2, column CAT H

mean

), we visualize

vector data using box plots (Figure 6), and are going

to use parametric unpaired two samples T-test to get

statistical evidence on the difference of data.

From the output (Open edX p-value = 0.2542, and

Moodle p-value = 0.4058), the two p-values are gre-

ater than the signiﬁcance level 0.05 implying that the

distribution of the data are not signiﬁcantly different

from the normal distribution. Here, we can assume

CSEDU 2018 - 10th International Conference on Computer Supported Education

510

Moodle Open edX

Cumulative Activity Time [hours]

Figure 6: CAT H - Cumulative Activity Time [hours] repre-

sented as second and third quartiles of both LMS.

the normality. Do the both CATH

mean

data sets have

the same variances? We use F-test to examine for

equality in data variances. The result is p-value =

7.839e-05. This result lead us to decline t-test met-

hod to compare data sets.

As an alternative, we use Wilcoxon rank sum test

for testing the equality of two distributions. We have

following results. We can conclude that CAT H

median

of Moodle LMS is signiﬁcantly different from Open

edX CAT H

median

with a p-value = 0.006051.

We also test, whether CATH(Moodle) is less than

the median CATH(Open edX). The result is: Open

edX CAT H

median

is greater than Moodle CAT H

median

The visual observation of the box-plot (Figure 6)

also clearly show: (1) the signiﬁcant difference bet-

ween CAT H data of Moodle and Open edX, and (2)

notable inequality in data variances.

4.3 Comparing Distributions

Two following ﬁgures (Figure 7 and Figure 8) give the

simple insight into the statistical data difference bet-

ween the Moodle and Open edX data using the histo-

grams of the CATH – Cumulative Activity Time ex-

pressed in Hours. The analysis shows that median and

mean average activity of users in Open edX is approx-

imately per 50 % better comparing to Moodle data.

Cumulative Activity Time per User [Hours]

Frequency

0 2 4 6 8 10

0 10 20 30 40 50 60

MSL < 40 min

MIN = 0

MAX = 29.2

MEAN = 3.5

MEDIAN = 2.7

Open edX, MSL < 40 [min]

MIN = 0

MAX = 29.2

MEAN = 3.5

MEDIAN = 2.7

Open edX, @MSL = 40 [min]

MIN = 0

MAX = 29.2

MEAN = 3.5

MEDIAN = 2.7

Figure 7: Open edX. CATH Histogram.

Cumulative Activity Time per User [Hours]

Frequency

0 2 4 6 8 10

0 10 20 30 40 50 60

Moodle, @MSL = 40 [min]

MIN = 0

MAX = 11.2

MEAN = 2.3

MEDIAN = 1.8

Figure 8: Moodle course. CATH Histogram.

4.4 Clicks and Activity Time

Figure 9 and Figure 10 depict how the number of

produced events (simply clicks) relates to estimated

Cumulative Activity Time of the learner. On canvas,

each point depicts one learner.

By comparing of both ﬁgures, we identify that

clicking and time that spent in the system, corre-

late. The correlation coefﬁcients of produced logged

events and activity time periods are 0.8020837 (Open

edX), and 0.8931734 (Moodle). Since both values are

rather close to 1, we can conclude that the variables

are positively linearly related.

0 1 2 3 4 5

0 200 400 600 800

Cumulative Activity Time [hours]

Number of Clicks per User

MSL = 40 [min]

edX

Figure 9: Open edX. Number of Clicks produced by the

learner as a function of Cumulative Activity Time.

0 1 2 3 4 5

0 200 400 600 800

Cumulative Activity Time [hours]

Number of Clicks per User

MSL = 40 [min]

Moodle

Figure 10: Moodle. Number of Clicks produced by the le-

arner as a function of Cumulative Activity Time.

Impact of LMS Selection on Students’ Activity

511

5 CONCLUSIONS

In general, only the one speciﬁc LMS can be analy-

zed efﬁciently using the proposed CAST algorithm.

The wide range of uncontrollable factors (samples

from two different students’ groups, dissimilar in-

terfaces, etc.) signiﬁcantly reduce the precision of

quantitative detected user behavior difference in both

LMS. Despite the impact of uncontrollable factors,

the CAST algorithm reduces the overestimation of the

learners’ activity.

After the application of the CAST algorithm to

LMS data, we expect to work further in the derivation

of the ”machine learning” models, useful for LMSs’

automation and dynamic adaptation to students’ and

teachers’ needs.

The trend to use Open edX more versus Moodle

can be easily veriﬁed by visual data analysis. This

lead to the categorical conclusion that students in the

Open edX environment in comparison with Moodle

LMS are more active.

Some other, not discussed in the paper beneﬁts of

Open edX (as Python assessments coding or LaTeX

formulae writing options) implies more engaged stu-

dents, and better-trained employees in the future. This

is the strategic business decision.

ACKNOWLEDGEMENTS

This research has been supported by a grant from

the international European ERA-NET Project Futu-

rICT2.0 funded under the FLAG-ERA Joint Transna-

tional Call 2016.

REFERENCES

Benjamin, A. (2011). Successful Remembering and

Successful Forgetting: A Festschrift in Honor of Ro-

bert A. Bjork. Taylor & Francis.

Bjork, R. (1994). Memory and meta-memory considerati-

ons in the training of human beings.

Cennamo, K. S. (2012). Learning from Video, pages 1876–

1879. Springer US, Boston, MA.

Coppola, G. and O’Higgins, N. (2015). Youth and the Cri-

sis: Unemployment, Education and Health in Europe.

Routledge Studies in Labour Economics. Taylor &

Francis.

Cornelius-White, J. H. D. and Motschnig, R. (2012).

Person-Centered Learning, pages 2596–2599. Sprin-

ger US, Boston, MA.

DeGroot, M. and Schervish, M. (2013). Probability and

Statistics: Pearson New International Edition. Pear-

son Education Limited.

Ifenthaler, D. and Seel, N. M. (2012). Stochastic Models

of Learning, pages 3192–3194. Springer US, Boston,

MA.

Kapenieks, K. (2013). ebig3 system portal, available at:

http://ebig3.eu/v2/.

Kumar, V. and Reinartz, W. (2006). Customer relationship

management: a database approach. Wiley.

Maniar, N. J. (2012). Video-Based Learning, pages 3403–

3405. Springer US, Boston, MA.

Mosteller, F. (1958). Stochastic models for the learning pro-

cess. Proceedings of the American Philosophical So-

ciety, 102(1):53–59.

Okoshi, T., Ramos, J., Nozaki, H., Nakazawa, J., Dey,

A. K., and Tokuda, H. (2015). Reducing users’ per-

ceived mental effort due to interruptive notiﬁcations

in multi-device mobile environments. In Proceedings

of the 2015 ACM International Joint Conference on

Pervasive and Ubiquitous Computing, UbiComp ’15,

pages 475–486, New York, NY, USA. ACM.

Pijeira D

ıaz, H. J., Santoﬁmia Ruiz, J., Ruip

erez-Valiente,

J. A., Mu

noz Merino, P. J., and Delgado Kloos, C.

(2016). A demonstration of analyse: A learning ana-

lytics tool for open edx. In Proceedings of the Third

(2016) ACM Conference on Learning @ Scale, L@S

’16, pages 329–330, New York, NY, USA. ACM.

Romero, C., Ventura, S., and Garc

ıa, E. (2008). Data mi-

ning in course management systems: Moodle case

study and tutorial. Comput. Educ., 51(1):368–384.

Seel, N. M., editor (2012). Blended Online Learning, pages

465–465. Springer US, Boston, MA.

Sikos, L. (2011). Web Standards: Mastering HTML5, CSS3,

and XML. Apressus Series. Apress.

TULVING, E. and Craik, F. (2000). The Oxford Handbook

of Memory. Oxford Library of Psychology Series. Ox-

ford University Press.

Wickham, H. and Grolemund, G. (2016). R for Data

Science: Import, Tidy, Transform, Visualize, and Mo-

del Data. O’Reilly Media.

Zumel, N. and Mount, J. (2014). Practical Data Science

with R. Manning Publications Company.

CSEDU 2018 - 10th International Conference on Computer Supported Education

512