Supporting Institutional Awareness and Academic Advising using

Clustered Study Proﬁles

Mariia Gavriushenko, Mirka Saarela and Tommi K¨arkk¨ainen

Faculty of Information Technology, University of Jyv¨askyl¨a, Jyv¨askyl¨a, Finland

Keywords:

Academic Advising, Learning Analytics, Robust Clustering.

Abstract:

The purpose of academic advising is to help students with developing educational plans that support their

academic career and personal goals, and to provide information and guidance on studies. Planning and man-

agement of the students’ study path is the main joint activity in advising. Based on a study log of passed

courses, we propose to use robust, prototype-based clustering to identify a set of actual study path proﬁles.

Such proﬁles identify groups of students with similar progress of studies, whose analysis and interpretation

can be used for better institutional awareness and to support evidence-based academic advising. A model of

automated academic advising system utilizing the possibility to determine the study proﬁles is proposed.

1 INTRODUCTION

The credit-based system is used to characterize the

requirements and progress of a student in many learn-

ing environments. Availability of personalized sup-

port is very important constituent of a learner’s suc-

cess (Nguyen et al., 2008). Academic advising (AA)

is an iterative collaboration process between student,

academic adviser, and academic institution, to tackle

the student retention. Advisers provide versatile as-

sistance to the students during their studies, making

the educational experience relevant and supported.

AA activity has a long history dating back to

1870s (Tuttle, 2000). Advising starts, when a student

becomes enrolled in higher education, and ﬁnishes

when the degree has been completed. The purpose of

academic advising is to ensure that the students carry

out the required studies to graduate. The central activ-

ity for this purpose is to support study planning, espe-

cially at the beginning of the academic life. Depend-

ing on the organizationalculture, especially the stabil-

ity or dynamicity of the course schedule, an academic

adviser either needs to ensure that predeﬁned study

plan is being followed or that a student knows all the

relevant study possibilities. Email, social media, web

and wikipages etc. provide means to share the nec-

essary information with the students. However, typ-

ically face-to-face discussions take place either regu-

larly (e.g., at the beginning of a semester or academic

year) or by students’ or advisers’ request. For more

personalized support, an adviser should know when

a student is in need of a study advice discussion and

what is the precise status of the studies.

Preliminary recommendations to do certain, espe-

cially compulsory, courses to proceed normally with

the major subject studies are provided by departments

and advisers. More precisely, for example at the Uni-

versity of Jyv¨askyl¨a (JYU), all students are required

to prepare an electronic personal study plan with the

academic adviser from the home department. Advis-

ing is organized according to the satellite model re-

ferring to the distributed responsibility of academic

units (Tuttle, 2000). The engagement model between

advisee and adviser characterize the principal spirit of

counselling (Feghali et al., 2011).

In JYU, the study plan and the completed studies

create the starting point for a study plan assessment

discussion. However, especially in computer science,

the actual number of studies that have been made dur-

ing an academic year are typically less than recom-

mended (Saarela and K¨arkk¨ainen, 2015a). Hence, it is

very common that the actual study path deviates from

the recommendations and plans, and in such a case an

advising intervention is needed. But how does an in-

dividualstudent and especially an individualstudy ad-

viser know, what is the relation between students’ re-

alized study path and that of the peer students? Thus,

could we, instead of comparing against the predeﬁned

plans, advise students based on evidence from the ac-

tual study paths of other similar students?

Therefore, the purpose of this article is to pro-

pose using a learning analytics method (Chatti et al.,

Gavriushenko, M., Saarela, M. and Kärkkäinen, T.

Supporting Institutional Awareness and Academic Advising using Clustered Study Proﬁles.

DOI: 10.5220/0006252300350046

In Proceedings of the 9th International Conference on Computer Supported Education (CSEDU 2017) - Volume 1, pages 35-46

ISBN: 978-989-758-239-4

2012), more precisely robust clustering (

Ayr¨am¨o,

2006; Saarela and K¨arkk¨ainen, 2015a), to create

groups of actual proﬁles of students concerning their

studies. Such proﬁles summarize the different typ-

ical accumulations of completed studies, increasing

the general awareness of the common study ﬂows.

One can explicitly link an individual student to stu-

dent peers with similar study path proﬁle in the same

institutional environment. This allows an adviser to

plan a possible intervention adaptively for a larger

pool of students instead of following each one indi-

vidually.

Creating general study proﬁles of students can

help departments in their assessment and planning

of when (and how) they provide the courses, espe-

cially the compulsory ones (Saarela and K¨arkk¨ainen,

2015a). By automating general student proﬁling it

is possible to provide essential support for adaptive

on-line advising along the lines suggested, e.g., in

(Nguyen et al., 2008; Henderson and Goodridge,

2015). Individual student’s perspective, from self-

regulated learning and study planning point of view

without academic advising interface, was thoroughly

addressed in (Auvinen, 2015) (see also (Auvinen

et al., 2014)).

The contents of this article is as follows: after the

introduction, we provide background on academic ad-

vising and personalized student support in Section 2.

Then, in Section 3, we describe the data and the ro-

bust clustering method, and introduce three cases to

construct student proﬁles to support academic advis-

ing. In Section 4 features and a model of an Academic

Adviser system with automated mechanisms in it are

proposed. The work is concluded in Section 5.

2 BACKGROUND

Academic advising is a collaborativeprocess in which

adviser and advisee enter a dynamic relationship

where adviser helps advisee to enhance the learn-

ing experience by helping in making academic deci-

sions (Henderson and Goodridge, 2015). The deci-

sion support could be made by analysis of student’s

records, as well as some external factors like interests,

goals, academic capabilities, schedules etc. (Noa-

man and Ahmed, 2015). Developmental Advising

means helping students to deﬁne and explore aca-

demic and career goals and pathways, as well as to

develop problem-solving and decision-making skills;

Prescriptive Advising, which is the more traditional

advising model, is mainly concentrated on providing

the information to the students according to their aca-

demic program, progress, academic policies, course

selection, etc.; Intrusive Advising refers to contact-

ing with student in critical periods like ﬁrst year of

study before the declaring major, graduation period,

or when students are at-risk or theyare high-achieving

students (Noaman and Ahmed, 2015).

Next we brieﬂy summarize a pool of directly

AA related work that was identiﬁed through a non-

systematic search. Our main concern here is to illus-

trate the strong link between the needs and practices

of AA and the general utility of student proﬁling.

2.1 On Academic Advising

Student voices on AA were raised in only some ar-

ticles. El-Ansiri et al. (Al-Ansari et al., 2015)

used questionnaire to study student satisfaction and

support-seeking patterns among dental students in

Saudi-Arabia. Very low (only 7.6%) primary util-

ity help rate of advisers in the academic matters was

encountered. Even if the advisers were available

when needed, they were not able to provide the most

relevant information, e.g., on important dates and

courses. Hence, up-to-date course and timetable in-

formation seems to be a prerequisite for AA, which is

handled by the in-house developed, integrated study

information system

Korppi

The pedagogical side of AA was also focused

rarely. In the work (Drozd, 2010) author studied aca-

demic advisers through the lens of transformational

leadership, i.e. how advisers can create a connec-

tion to students that positively inﬂuence their study

paths (by increasing and inspiring study motivation

and engagement/commitment in studies through in-

dividual and intellectual consideration). A question-

naire for undergraduate students strengthened the im-

portance of transformational leadership activities in

adviser-student communication and collaboration, in-

dependently from the student’s characteristics. The

lack of time for individual counselling efforts that was

visible in most of the reviewed articles here was not

emphasized in (Drozd, 2010). Dougherty (Dougherty,

2007) studied academic advisers from those students’

perspective who are doing very well in their stud-

ies. These students are called high-achieving stu-

dents. Authors address the need for the investigation

of unique characteristics of these students.

Technical support for AA has been considered in

many articles. The availability of extensive infor-

mation on courses to support automatization of AA

was emphasized in (Biletskiy et al., 2009). The au-

thors proposed course outline data extractor applica-

tion, which helps in recognizing similar or compara-

ble courses between different institutions, also help-

https://www.jyu.ﬁ/itp/en/korppi-guide

CSEDU 2017 - 9th International Conference on Computer Supported Education

ing both students and academic advisers to keep track

of the variety of topics that i) have been covered in the

completed studies, ii) should be covered to complete

minor or major subject modules or the actual degree.

The authors in (Nguyen et al., 2008) proposed an

integrated knowledge-based framework based on se-

mantic technology that supports computer-based (au-

tomatic) e-Advising on the suitable courses for the

students. Naturally individual learning history data

provide the starting point for the system and, for this

purpose, the authors implemented and tested a data

integration tool.

The high workload of academic advisers, espe-

cially due to individual but many times recurrent han-

dling of basic issues with multiple students in a hurry,

was addressed in (Henderson and Goodridge, 2015),

with the proposition of an intelligent, semantic, web-

based application to assist decision making and au-

tomatization of repetitive counselling tasks. Core of

the system consisted of rule-based inference engine,

which mapped student proﬁle with the study pro-

gram proﬁle and organizational rules, to provide au-

tomatic suggestions on the courses to be enrolled in

the upcoming semester. In the preliminary evaluation,

a positive feedback of the system was obtained, al-

though the main limitation of suitability to only study

programs which follow a clear, predeﬁned study path

of courses, was recognized. With very similar aims

and functionality, another web-based on-line adviser

was described in (Feghali et al., 2011). This system

was also evaluated positively when compared to the

current advising system. The authors emphasized that

such a tool only supports and does not replace a hu-

man academic adviser.

Conversational, fully autonomous agent support-

ing AA dialogs using natural language processing

(NLP) were suggested in (Latorre-Navarro and Har-

ris, 2015). The proposed system contained an exten-

sible knowledge base of information and rules on aca-

demic programs and policies, course schedules, and

a general FAQ. NLP performance of the proposed

system was evaluated positively. Also, the similar

multi-agent approach was suggested in (Wen and Mc-

Greal, 2015) for AA. This approach helps tackling a

dynamic and complex individualized study planning

and scheduling problem. As well as in (Al-Sarem,

2015) was proposed a decision tree model for AA

affairs based on the algorithm C4.5. The output is

evaluated based on Kappa measure and ROC area.

The main conclusion was made that the difference be-

tween the registered and gained credit hours by a stu-

dent was the main attribute that academic advisers can

rely on (Al-Ansari et al., 2015).

As can be concluded, earlier studies have mostly

concentrated on research prototypes which focus only

on few main components or tool support for existing

learning management systems. Taking into account

that user modeling is one of the key factors for includ-

ing personalization into the learning system, many re-

searchers used ontologies for learners’ models, be-

cause ontologies have many advantages for creation

of user models (Idris et al., 2009; Chen, 2009a; Le-

ung et al., 2010; Nguyen et al., 2008; Biletskiy et al.,

2009; Henderson and Goodridge, 2015).

Data-mining techniques have also been applied to

the learning environments in order to track users’ ac-

tivities, extract their behavior proﬁles and patterns,

and analyze the data for future improvement of the

learning results, as well as for identifying types of

learners (Minaei-Bidgoli, 2004). Mostly, for develop-

ing personalized learning plan, researchers used deci-

sion tree search, heuristic algorithms, genetic algo-

rithms, item response theory and association rules.

Also, many studies used semantic web technologies,

neural networks and multi-agent approach. Most of

the previous studies on personalized learning path

generation schemes have mainly focused on guiding

the students to learn in the digital world; i.e. each

learning path represents a set of digitalized learn-

ing objects that are linked together based on some

rules or constraints (Liu et al., 2008). While deter-

mining such digitalized learning paths, the learning

achievements, on-line behaviors or personalized fea-

tures (such as learning style) of individual students

are usually taken into consideration (Schiafﬁno et al.,

2008; Chen, 2008; Chen et al., 2008; Chen, 2009b;

Chen et al., 2005).

2.2 On Personalization of Student

Support

In general, many researchers have paid attention to

developing e-learning systems with personalization,

and the most common aspect in these system is the

creation of the personalized learning path for each

individual student or group of students. Most of per-

sonalized systems consider learner preferences, in-

terests and browsing behaviors, because it will help

to provide personalized curriculum sequencing ser-

vice (Huang et al., 2007). In the study (Chen et al.,

2005) authors proposed a personalized e-learning sys-

tem which is based on Item Response Theory (PEL-

IRT). This system is considering course material dif-

ﬁculty and learner ability, to provide individual learn-

ing path for learners. Learner’s ability estimation was

based on an explicit learner’s feedback (the answers

of learners to the assigned questionnaires). The sys-

tem appeared mostly like a recommendation system

Supporting Institutional Awareness and Academic Advising using Clustered Study Proﬁles

of the courses for the learners. Authors in (Huang

et al., 2007) proposed a genetic-based curriculum se-

quencing approach and used case-based reasoning to

develop a summative assessment. The empirical part

indicated that the proposed approach can generate ap-

propriate course materials for learners taking into ac-

count their individual requirements. Later, in (Chen,

2009a), the authors developed a personalized web-

based learning system grounded on curriculum se-

quencing based on a generated ontology-based con-

cept map, which was constructed by the pre-test result

of the learners. Optimization problem for modeling

criteria and objectives for automatic determination of

personalized context-aware ubiquitous learning path

was suggested in (Hwang et al., 2010). This learn-

ing model not only supports learners with alternative

ways to solve problems in real-world situations, but

also proposes more active interaction with the learn-

ers. Authors in (Werghi and Kamoun, 2009) proposed

Decision Support System for student advising based

on decision tree for an automated program planning

and scheduling. The proposed approach takes into

account prerequisite rules, the minimum time (mini-

mum number of terms), and the academic recommen-

dations. The adaptive course sequencing for personal-

ization of learning objectives was suggested in (Idris

et al., 2009) using neural networks, self organizing

maps and the back-propagation algorithm.

A very closely related work to ours was reported

in (Sandvig and Burke, 2005). Authors proposed

a case-based reasoning paradigm which is based on

the assumption that similar students will have simi-

lar course histories. The system used the experience

and history of graduated students in order to propose

potential courses for the students. Unfortunately, this

approach required matching between students’ histo-

ries. Also, similar case-based reasoning was used by

(Mostafa et al., 2014) for developing a recommen-

dation system for a suitable major to students based

on comparison of the student information and similar

historical cases.

As reviewed, many suggestions for intelligent

software and information system support of AA have

been given. Many studies describe the creation of

intelligent learning systems that can make a curricu-

lum sequencing more ﬂexible for providing students

with personalized and adaptive study support services

(Fung and Yeung, 2000; Lee, 2001; Brusilovsky,

1998; Lee, 2001; Papanikolaou et al., 2002; Tang

and McCalla, 2005). Universities are more and more

looking into developing self-service systems with in-

telligent agents as an addition or replacement for the

labor-intensive services like academic advising. For

example, The Open University of Hong Kong has de-

veloped an intelligent on-line system that instantly re-

sponds to enquiries about career development, learn-

ing modes, program/course choices, study plans, and

graduation checks (Leung et al., 2010).

However, the institutional starting point concern-

ing available digital information, especially for the

web-based systems that have been proposed, seems

to vary a lot. Some systems start and focus on pro-

viding easy access to course and degree requirements

information whose availability is to be assured ﬁrst.

On the other hand, we might start from the situa-

tion where we can readily access most of the rele-

vant data: i) course information with basic contents,

learning goals, assessment methods, acceptance crite-

ria, schedule and location, teachers and lecturers etc.;

ii) individual, anonymous study records on passed

courses and completed studies. (Note that reliable in-

formation on student admission is currently not di-

rectly available in the organization under considera-

tion).

3 CREATION OF STUDY PATH

PROFILES USING ROBUST

CLUSTERING

3.1 Data

To illustrate the proposed approach, we utilized real

study records of the Bachelor (BSc) and Master

(MSc) students majoring in Mathematical Informa-

tion Technology (which is comparable to a major in

Computer Science at other universities) at the Uni-

versity of Jyv¨askyl¨a (JYU/MIT). IT administration at

the University has recently created a data warehouse

of passed courses by all the students, which can be

utilized by the departments. On the other hand, the

electronic study plan system does not provide direct

interface for larger student groups, so both from ac-

cessibility and evidence-basedness points of view, we

focus on analyzing the real study log of the passed

courses. The log was anonymized, keeping student

IDs as keys, covering the four calendar years 2012–

2015. Note that students can start their studies in

the beginning of September (autumn term) or January

(spring term). Hence, the original study registry log

included a heterogeneous set of BSc and MSc stu-

dents who had started their studies either before 2012

or in the beginning of spring or autumn terms during

2012 – 2015.

The whole study log contained 15370 passed

courses by 1163 different students on 1176 different

course IDs. There were 942 male students (81%),

CSEDU 2017 - 9th International Conference on Computer Supported Education

Figure 1: Probability of size of a course.

with mean amount of studies made 59.9 ECTS. Only

221 female students (19%) were identiﬁed, with mean

amount of studies made 57.0 ECTS. Hence, most of

the students in the log were either in the beginning of

their studies or progressing very passively and slowly.

Figure 1 shows the discrete density distribution

of the size of the passed courses. According to the

ﬁgure, 5 ECTS and 3 ECTS are the two most com-

mon sizes of the courses, the former covering around

30% of the studies. Moreover, there are a lot of small

courses (1 – 6 ECTS) with the exception of the MSc

thesis, 30 ECTS. Teaching in JYU is organized for

four periods during one academic year (plus the sum-

mer semester) in such a way that a course of ca. 5

ECTS can ﬁt to one period. We conclude that because

the passed courses represent both major and minor

subject studies, division of the overall learning ob-

jects as courses is not optimal. This observation is

the ﬁrst example on how summarization of study log

data provides visibility and feedback to the organi-

zation. During the course of writing this article, we

also found out that the instructions of JYU for prepar-

ing the next curriculum for 2017–2019 include strong

recommendation to decrease the number of courses

with only a few credits.

Next we aggregated how many credits per

semester each student had made. Similarly to (Saarela

and K¨arkk¨ainen, 2015a), each calendar year was di-

vided into two semesters: the spring term (from Jan-

uary to June) and the autumn term (from July to De-

cember). However, since usually only a few courses

are completed during the summer (this is illustrated

in (Saarela and K¨arkk¨ainen, 2015a)), it was reason-

able to divide the calendar year into only two parts

for further analysis.

In what follows, we proﬁle, analyse and compare

two students cohorts: those who started their studies

in the beginning of the autumn term 2012 (A2012) or

2013 (A2013). Hence, for A2012 we end up with 8

and for A2013 with 6 integer variables representing

the aggregated amount of credits on half-a-year scale.

Since the students have progressed in their studies

very differently and many of them have not been ac-

tive during all the semesters of interest, both of the

data sets are very sparse containing a lot of missing

values (Saarela and K¨arkk¨ainen, 2015a). This is the

key property that is taken into account in the proﬁling

approach that is described next.

3.2 Robust Clustering Method

As already explained, our goal is to assist the aca-

demic advisers by recommending suitable courses for

students based on passed courses of (possibly more

advanced) students with similar study path. For this,

we need to identify general proﬁles of similar students

and this is, precisely, the purpose of clustering. Par-

titional (or representative-based (Zaki and Meira Jr,

2014)) clustering seems to be the right family of clus-

tering methods to choose from because it assigns each

observation to exactly one cluster, which is repre-

sented by its most characteristic point, the cluster cen-

troid, which represents the common proﬁle. Within a

cluster, distances of observations to the prototype de-

termine the most typical or representative members of

a cluster. Thus, instead of following many different

student proﬁles, the academic adviser can just follow

the most common proﬁles to get an overview of the

whole cohort.

Generally, partitional based clustering algorithms

consist of an initialization step, in which the initial

centroids of each cluster are generated, and iterations

of two steps where (i) each observation is assigned

to its closest centroid, and (ii) the centroid of each

cluster is recomputed by utilizing all observations as-

signed to it. The algorithm stops when the centroids

remain the same over two iterations. The most pop-

ular and most applied partitional clustering algorithm

is the k-means (Jain, 2010), also in learning analytics

studies (Saarela and K¨arkk¨ainen, 2017). This algo-

rithm works very well for full and approximately nor-

mally distributed data since the sample mean is the

most efﬁcient estimator for samples that are drawn

from the normal distribution. However, the sample

mean is highly sensible to all kinds of outliers (Huber,

2011) as well as missing values, which can be char-

acterized as special types of outliers. Also for a non-

symmetric (skewed) distribution, the sample mean is

not necessarily the most efﬁcient estimator and other

location estimates might be preferable (Sprent and

Smeeton, 2016). Moreover, as explained in (Saarela

and K¨arkk¨ainen, 2015a), the quantization error for the

Supporting Institutional Awareness and Academic Advising using Clustered Study Proﬁles

integer-type variables like here has uniform not gaus-

sian distribution.

The spatial/geometric median is a robust nonpara-

metric location estimate, which remains reliable even

if half of the data is contaminated (Sprent and Smee-

ton, 2016). Mathematically, the spatial median is the

Weber point that minimizes the (nonsquared) sum of

the Euclidean distances to a group of given points

}, i = 1, 2, . . . n:

argmin

∑

i=1

− ck.

Although the basic concept is easily understood and

has been extensively discussed in the literature (albeit

under various names, see (Drezner and Hamacher,

2001)), its computation is known to be difﬁcult.

In (

Ayr¨am¨o, 2006), the difﬁculty of computing

the spatial median during partitional clustering was

solved with the SOR (Sequential Overrelaxation) al-

gorithm (see (

Ayr¨am¨o, 2006) for details). More-

over, in the implementation of the resulting k-spatial-

medians clustering algorithm, only the available (i.e.

not-missing) data is taken into account when the cen-

troid is recomputed.

To sum up, all of these above discussed prop-

erties – most importantly, the robustness to missing

data and the fact that every cluster is represented by a

centroid – make the k-spatial-medians clustering very

suitable for creating student’s general study proﬁles.

The fact that such a clustering approach works very

well for sparse educational data has been previously

shown in (Saarela and K¨arkk¨ainen, 2014; Saarela and

K¨arkk¨ainen, 2015b; Saarela and K¨arkk¨ainen, 2015).

The initialization of the robust clustering method

was realized similarly as in (Saarela and K¨arkk¨ainen,

2015): We started with multiple repetitions of k-

means for the complete data – without missing val-

ues – and then, applied k-spatial-medians to the best

of those results.

3.3 Clustered Student Proﬁles

Similarly to the earlier work in (Saarela and

K¨arkk¨ainen, 2014; Saarela and K¨arkk¨ainen, 2015b;

Saarela and K¨arkk¨ainen, 2015a; Wallden, 2016), we

apply four different internal cluster validation indices

to determine the number of clusters: Knee Point

(KP) of the clustering error, Ray-Turi (RT), Davies-

Bouldin (DB), and Davies-Bouldin∗ (DB∗). All the

computations here were carried out in the Matlab-

environment, using own implementations of all the

algorithms.

From the two student groups A2012 and A2013,

we include in clustering only still active students, i.e.

those who have made credits during the autumn term

2015 (the last one analyzed). Furthermore, we restrict

ourselves to those students for whom over half of the

variables are available (Sprent and Smeeton, 2016).

This means that the 47 analyzed students in A2012

have made studies during at least four out of the seven

possible semesters (including the last one) and the

76 students in A2013 at least in three out of the ﬁve

semesters. Because of the anonymity, we obtained

further assistance in relation to the metadata and inter-

pretation of the clusters from the Study Amanuensis

of the Department (Study Amanuensis, 2016).

Figure 2: Boxplot for A2012.

Figure 3: Credit accumulation prototypes for A2012.

A2012

The boxplot in Figure 2 shows the large variability

in the study accumulations both within semesters and

between semesters. We see the larger accumulations

in the spring terms during the ﬁrst two years, and

a slightly decreasing overall trend after that. There

are always exceptional students who have made much

more studies than their peers.

KP, DP, and DP∗ indicated four clusters and RT

had also local minimum there, so we choose to an-

alyze four different general study progress proﬁles.

The proﬁles for A2012 are depicted in Figure 3,

CSEDU 2017 - 9th International Conference on Computer Supported Education

where the size of the cluster is given in the top-right

corner. The proﬁles are sorted in the ascending order

with respect to the total number of credits.

The main group of 21 students in the ﬁrst clus-

ter illustrate a potential start of the studies in the ﬁrst

year, with strong passivation after that. They have ob-

tained prototypically 65 ECTS until the end of 2015.

Based on (Study Amanuensis, 2016), by a closer look

on the 8 students from the cluster closest to the cen-

troid, these are all older BSc and MSc male students

(born before 1990). They are either distant students

studing while working or have completely chosen to

change their orientation from an earlier occupation

and already ﬁnished degree. The difﬁculties in studies

and reasons of such a behavior, for a similar adult stu-

dent proﬁle, were thoroughly discussed in the earlier

work (Kaihlavirta et al., 2015) from the same context

(department) than here.

The second group of only 7 students, who gen-

erally obtained 103 ECTS, shows opposite behavior:

very slow start in the ﬁrst year, activating to an appro-

priate level then. Three most characteristic students

here were young males, who were involved in the mil-

itary service during the ﬁrst study year. This complete

explains the observed behavior.

The third group of 10 students, who generally ob-

tained 147 ECTS, did their studies very actively for

the ﬁrst 4–5 semesters. Analysis of the three most

characteristics students revealed two young and one

older male students who either took job or became

active in student organizations during the third year

of the studies.

The fourth proﬁle with 9 students, altogether 184

ECTS in general, illustrates that a good start on

the study activity carries over the semesters. Three

mostly characteristics students were again all males,

one MSc student and two BSc students. Note that

similar ﬁnding on the importance of active start in

an individual course level was given in (Saarela and

K¨arkk¨ainen, 2015a).

Students who are mostly in need of academic ad-

vising are the ones in the ﬁrst cluster. They can be

identiﬁed either in the beginning of their studies or af-

ter the second semester, because even if still making

studies, their accumulation is much less than in the

third and fourth cluster. Their characterization also

suggests the department to rethink the study entrance

criteria.

A2013

For A2013 all cluster indices suggested three pro-

ﬁles, which are illustrated in Figure 5. This and the

fact that there are now one proﬁle less than in A2012

suggests more stable organization of the curriculum.

Figure 4: Boxplot for A2013.

Figure 5: Credit accumulation prototypes for A2013.

Also the boxplot in Figure 4 supports such ﬁnding,

especially showing smaller variability in the obtained

credits between the autumn and spring terms com-

pared to A2012 in Figure 2.

Student group in the need of intrusive academic

advising consists of those 23 students with small-

est accumulation of credits. These students start and

continue very slowly in their studies, although the

level of activity was increasing in the fourth and ﬁfth

semesters. Their general ECTS accumulation after

ﬁve semesters was 44 ECTS. Analysis of the ﬁvemost

representative students revealed two older male stu-

dents (birth year before 1990), two males with indi-

cations of military service, and a female student. Ac-

cording to (Study Amanuensis, 2016), especially the

younger students showed signs of low self-regulation

during academic advising sessions.

The second proﬁle of 17 students, completing typ-

ically 80 ECTS, showed similar behavior to the sec-

ond proﬁle in A2012: the minimal ﬁrst year is raised

to a good level of study activity later. A closer look on

the ﬁve most representative students showed young,

two female and three male students. Four of these had

identiﬁed themselves as a non-active student during

the ﬁrst study year, again mostly due to the military

service of the young male students.

The third proﬁle of 36 most active students, ac-

Supporting Institutional Awareness and Academic Advising using Clustered Study Proﬁles

complishing 123 ECTS typically, showed similar

overall behavior than the fourth proﬁle in A2012. The

ﬁrst semester is slightly smaller but then the study

path proceeds in the desired way. Recapitulation of

the meta data of ﬁve most representative students

showed ﬁve male students, of whom three were ori-

ented towards game programming and development -

the most recent study line of the department.

We note that even if the boxplot in Figure 4 in-

dicated more stable study path with respect to au-

tumn and spring semesters, the two proﬁles of truly

active students still illustrated larger study accumula-

tions in the spring than in the autumn. These ﬁndings

are, however, mostly explained by the longer calen-

der time for the two periods in the spring term com-

pared to the autumn term – a general peculiarity of the

Finnish higher education system.

The similarities and differences between the two

sets of proﬁles just discussed emphasize the impor-

tance of the use of evidence-based information in aca-

demic advising. On one hand, there are repetitive pro-

ﬁles of students proceeding in their studies well or

slowly. The latter ones needs to be detected and sup-

ported in an intrusive manner in academic advising.

The home department responsible for major subject

studies and the other departments providing minor

subject studies should be informed about the found

hindrances of the study paths. In the case analyzed

here, there is a clear change of study accumulation

proﬁles from A2012 to A2013, which suggests that

the organization of courses, the capabilities of stu-

dents, and/or their support through academic advising

have improved in the educational organization under

study.

4 PROPOSITION OF A SYSTEM

MODEL

As shown, it is important to follow the actual progress

of the students in their studies. There might be no

need for an advising intervention, but if so, one should

automatically notify the students and the study coun-

selling on the deviations in the study path. The

problems of not passing courses and not following

suggested study plans usually also call for organiza-

tional considerations whether learner ability and the

difﬁculty level of the recommended curriculum are

matched to each other properly (Huang et al., 2007).

4.1 Proposed System Architecture

This subsection describes the novel system architec-

ture for AA and automatic feedback based on recog-

nized student group proﬁles which are obtained by us-

ing clustering. We also present an overviewof the AA

process as both manual and automated process.

The architecture of the proposed system’s model

for the AA is presented in Figure 6. The system

has two main databases: learner proﬁle database

and curriculum database. Learner proﬁle database

stores learner’s data about studies, assessment re-

sults, timetables of completed studies, etc. Curricu-

lum database stores information about compulsory

courses, other courses, timetables, etc. The academic

advising system’s part consists of several blocks like

linking individual students to their peers with similar

study path proﬁle together with the recommendation

block and planning block.

Based on the system architecture, the details of

system’s main functionality read as follows:

1. Collection of learner’s personal information.

2. Collection of information about the courses and

completed studies.

3. Creation of study progress proﬁles along the lines

of Section 3.

4. Linking the individual student to student peers

with similar study proﬁle in the institutional en-

vironment.

5. Student’s progress check. If student is linked to a

proﬁle requiring intrusiveadvising, inform the ad-

viser and the student by providing the interpreted

study proﬁle to support the communication and

problem solving.

6. Modiﬁcation of the study plan on recommended

courses and their timetables by taking into ac-

count the evidence related to the identiﬁed study

proﬁle.

Figure 6: Architecture of the Academic Adviser.

CSEDU 2017 - 9th International Conference on Computer Supported Education

Figure 7: Automated Academic Advising process.

Figure 8: Comparison of the manual and automated pro-

cesses of learner’s study life cycle.

7. Planning and realization of an intrusive proﬁle in-

tervention adaptively for a larger pool of students.

Data collection related to the system is, naturally,

all the time active. The evidence-based study proﬁles

can and should be recomputed on regular basis. A

natural suggestion would be to do this after the studies

made during the previous semester have been stored

and become available for clustering.

4.2 Automated Academic Advising

Process

The automated process of Academic Advising, re-

lated to the system’s architecture and main function-

ality as described above, is presented in Figure 7. The

given proposition allows manual control of the contin-

uous advising activity for every learner individually,

or the more automated process where the role of the

advisers is shifted to the higher level of abstraction.

The difference on the level of learner’s life cycle be-

tween these two use cases is depicted in Figure 8. The

automated process is highlighted with the red color

in the ﬁgure. In the automated scenario, the respon-

sible persons of the study organization only provide

policies, planning and regulations. This can reduce

the responsibility for the daily routine work and could

help to provide recommendations for a larger pool of

students rather than for the each individual learner.

The work-ﬂow related to Figure 7 reads as fol-

lows: The learner is choosing the study program of an

educational institution. After that he or she chooses

with AA the proper courses which are related to the

Supporting Institutional Awareness and Academic Advising using Clustered Study Proﬁles

chosen program and creates a study plan. Information

about the student, the required courses and progress

in them is stored in the database and is automatically

changed/refreshed after each passed course. After

passing several courses, system can attach a student

to a group of students with similar, actual study path.

If learners are doing well, evidence-based determina-

tion and communication of this during advising en-

courages them to continue like that. If they are at-

tached to a proﬁle which does not progress with the

studies as expected, the system can identify this early

and provide intrusive academic advising support for

both the advisor and the students in question.

The proposed automated mechanism solve an im-

portant problem of improving and providing aca-

demic advising, because more and more students

should receive guidance with their study plans before

graduating. This system will help to plan when and

how to provide the courses, especially the compulsory

ones, as well as to plan a proﬁle intervention adap-

tively for a larger pool of students, which will reduce

the human effort of academic advising.

5 CONCLUSIONS

Academic Advising is an essential part of daily ac-

tivities in an educational institution and an important

component in the learner’s study life. Nowadays, we

need to be able to create and manage personalized

study plans and study paths taking into account learn-

ers abilities and regulations of the learning environ-

ment. And in order to better help students, Academic

Adviser should be able to manage a rich set of infor-

mation, e.g., on short-range program planning, evalu-

ation of students, and generation of the proper teach-

ing schedule, as well as plan possible interventions

adaptively for a group of students instead of follow-

ing all individual students separately. It is decisive

that learner should receive proper advising – poor or

no advising is known to have a negative effect on the

progress in studies (Al-Ansari et al., 2015).

In this paper,we presented a compact literature re-

view about Academic Advising, mostly focusing on

Automated Academic Advising and Intelligent Aca-

demic Advising. It was then described how, by using

a robust variant of prototype-basedclustering method,

which is especially suitable for data with missing val-

ues, one can create prototypical student group proﬁles

characterizing the overall progress of the studies. This

allows academic advisers to provide evidence-based

information on the study paths that were actually re-

alized by individual students. Moreover, academic in-

stitutions can focus on management and updates on

course schedule having an effect on clearly charac-

terized and recognized groups of students. Note that

even if the sample groups of students that were pro-

ﬁled here were very small, the used method is scal-

able to hundredsof thousands of students (Saarela and

K¨arkk¨ainen, 2015b).

Then a reference model for automated Academic

Advising system was proposed. The proposed archi-

tecture and model of the system are intended for a

development phase to prototype the whole automated

process, where the learners will be proﬁled regularly,

and where the proper study path will be presented,

as well as deviating learners detected. The proposed

model of the AA system will have automated process

of study path recommendation. This system will help

to plan when and how to provide the courses, espe-

cially the compulsory ones, as well as to plan a proﬁle

intervention adaptively for a larger pool of students,

which will reduce the human effort of academic ad-

vising.

By continuing the development of the line of

work, we could consider the study paths with higher

granularity than per semester. Also, the main func-

tionality of the proposed system – to provide an au-

tomated notiﬁcation for the academic advisers about

students and their progress, with the interpretation of

needs to modify and re-plan the study path – should

be properlyevaluated. Moreover, better availability of

learner’s personal information concerning the study

entrance criteria and current life situation, e.g., a part-

time job or living far from the institute, could support

both interpretation of the generated student proﬁles

and better preparation and management of the inter-

vention patterns of academic advising.

REFERENCES

Al-Ansari, A., El Tantawi, M., AbdelSalam, M., and Al-

Harbi, F. (2015). Academic advising and student

support: Help-seeking behaviors among Saudi den-

tal undergraduate students. The Saudi dental journal,

27(2):57–62.

Al-Sarem, M. (2015). Building a decision tree model for

academic advising affairs based on the algorithm C 4-

5. arXiv preprint arXiv:1511.04026.

Auvinen, T. (2015). Educational Technologies for Support-

ing Self-Regulated Learning in Online Learning Envi-

ronments. Aalto University.

Auvinen, T., Paavola, J., and Hartikainen, J. (2014).

STOPS: a graph-based study planning and curriculum

development tool. In Proceedings of the 14th Koli

Calling International Conference on Computing Ed-

ucation Research, pages 25–34. ACM.

Ayr¨am¨o, S. (2006). Knowledge Mining Using Robust Clus-

CSEDU 2017 - 9th International Conference on Computer Supported Education

tering, volume 63 of Jyv¨askyl¨a Studies in Computing.

University of Jyv¨askyl¨a.

Biletskiy, Y., Brown, J. A., and Ranganathan, G. (2009).

Information extraction from syllabi for academic

e-Advising. Expert Systems with Applications,

36(3):4508–4516.

Brusilovsky, P. (1998). Adaptive educational systems on the

world-wide-web: A review of available technologies.

In Proceedings of Workshop” WWW-Based Tutoring”

at 4th International Conference on Intelligent Tutor-

ing Systems (ITS’98), San Antonio, TX.

Chatti, M. A., Dyckhoff, A. L., Schroeder, U., and Th¨us,

H. (2012). A reference model for learning analytics.

International Journal of Technology Enhanced Learn-

ing, 4(5-6):318–331.

Chen, C.-M. (2008). Intelligent web-based learning system

with personalized learning path guidance. Computers

& Education, 51(2):787–814.

Chen, C.-M. (2009a). Ontology-based concept map for

planning a personalised learning path. British Jour-

nal of Educational Technology, 40(6):1028–1058.

Chen, C.-M. (2009b). Personalized e-learning system with

self-regulated learning assisted mechanisms for pro-

moting learning performance. Expert Systems with

Applications, 36(5):8816–8829.

Chen, C.-M. et al. (2008). Ontology-based concept map

for planning personalized learning path. In IEEE In-

ternational Conferences on Cybernetics & Intelligent

Systems.

Chen, C.-M., Lee, H.-M., and Chen, Y.-H. (2005). Person-

alized e-learning system using item response theory.

Computers & Education, 44(3):237–255.

Dougherty, S. B. (2007). Academic advising for high-

achieving college students. Higher Education in Re-

view, 4:63–82.

Drezner, Z. and Hamacher, H. W. (2001). Facility location:

applications and theory. Springer Science & Business

Media.

Drozd, D. S. (2010). Student preferences for academic ad-

visors as transformational leaders. PhD thesis, Texas

A&M University.

Feghali, T., Zbib, I., and Hallal, S. (2011). A web-based

decision support tool for academic advising. Educa-

tional Technology & Society, 14(1):82–94.

Fung, A. and Yeung, J. (2000). An object model for a web-

based adaptive educational system. In Proceedings

of the IFIP International Conference on Educational

Use of Technologies (ICEUT2000).

Henderson, L. K. and Goodridge, W. (2015). AdviseMe:

An intelligent web-based application for academic ad-

vising. International Journal of Advanced Computer

Science and Applications, 6(8):233–243.

Huang, M.-J., Huang, H.-S., and Chen, M.-Y. (2007). Con-

structing a personalized e-learning system based on

genetic algorithm and case-based reasoning approach.

Expert Systems with Applications, 33(3):551–564.

Huber, P. J. (2011). Robust statistics. Springer.

Hwang, G.-J., Kuo, F.-R., Yin, P.-Y., and Chuang, K.-

H. (2010). A heuristic algorithm for planning per-

sonalized learning paths for context-aware ubiquitous

learning. Computers & Education, 54(2):404–415.

Idris, N., Yusof, N., Saad, P., et al. (2009). Adaptive course

sequencing for personalization of learning path using

neural network. Int. J. Advance. Soft Comput. Appl,

1(1):49–61.

Jain, A. K. (2010). Data clustering: 50 years beyond K-

means. Pattern Recognition Letters, 31(8):651–666.

Kaihlavirta, A., Isom¨ott¨onen, V., and K¨arkk¨ainen, T.

(2015). A self-ethnographic investigation of contin-

uing education program in engineering arising from

economic structural change. Studies in Continuing

Education, 37(1):99–114.

Latorre-Navarro, E. M. and Harris, J. G. (2015). An intel-

ligent natural language conversational system for aca-

demic advising. Editorial Preface, 6(1).

Lee, M.-G. (2001). Proﬁling students’ adaptation styles

in web-based learning. Computers & Education,

36(2):121–132.

Leung, C. M., Tsang, E. Y., Lam, S., and Pang, D. C.

(2010). Intelligent counseling system: A 24 x 7 aca-

demic advisor. Educause Quarterly, 33(4):n4.

Liu, C.-l., Wu, S., Chang, M., and Heh, J. (2008). Guid-

ing students to do remedial learning in school campus

with learning objects’ spatial relations. In 16th inter-

national conference on computers in education con-

ference, pages 249–256. Citeseer.

Minaei-Bidgoli, B. (2004). Data mining for a web-based

educational system. PhD thesis, Michigan State Uni-

versity.

Mostafa, L., Oately, G., Khalifa, N., and Rabie, W. (2014).

A case based reasoning system for academic advising

in egyptian educational institutions. In 2nd Interna-

tional Conference on Research in Science, Engineer-

ing and Technology (ICRSET2014) March, pages 21–

22.

Nguyen, T. B., Nguyen, D. N., Nguyen, H. S., Tran, H.,

and Hoang, T. A. D. (2008). An integrated approach

for an academic advising system in adaptive credit-

based learning environment. Journal of Science, Nat-

ural Sciences and Technology, (24):110–121.

Noaman, A. Y. and Ahmed, F. F. (2015). A new framework

for e academic advising. Procedia Computer Science,

65:358–367.

Papanikolaou, K. A., Grigoriadou, M., Magoulas, G. D.,

and Kornilakis, H. (2002). Towards new forms of

knowledge communication: the adaptive dimension

of a web-based learning environment. Computers &

Education, 39(4):333–360.

Saarela, M. and K¨arkk¨ainen, T. (2014). Discovering

Gender-Speciﬁc Knowledge from Finnish Basic Ed-

ucation using PISA Scale Indices. In Proceedings of

the 7th International Conference on Educational Data

Mining, pages 60–68.

Saarela, M. and K¨arkk¨ainen, T. (2015a). Analysing Stu-

dent Performance using Sparse Data of Core Bachelor

Courses. JEDM-Journal of Educational Data Mining,

7(1):3–32.

Saarela, M. and K¨arkk¨ainen, T. (2015b). Do Country

Stereotypes Exist in PISA? A Clustering Approach for

Large, Sparse, and Weighted Data. In Proceedings of

the 8th International Conference on Educational Data

Mining, pages 156–163.

Supporting Institutional Awareness and Academic Advising using Clustered Study Proﬁles

Saarela, M. and K¨arkk¨ainen, T. (2015). Weighted Cluster-

ing of Sparse Educational Data. In Proceedings of

the 23rd European Symposium on Artiﬁcial Neural

Networks, Computational Intelligence and Machine

Learning (ESANN 2015), pages 337–342.

Saarela, M. and K¨arkk¨ainen, T. (2017). Knowledge Dis-

covery from the Programme for International Student

Assessment. In Pe˜na-Ayala, A., editor, Learning An-

alytics: Fundaments, Applications, and Trends: A

View of the Current State of the Art to Enhance e-

Learning, pages 229–267. Springer International Pub-

lishing, Cham.

Sandvig, J. and Burke, R. (2005). Aacorn: A cbr rec-

ommender for academic advising. Technical report,

Technical Report TR05-015, DePaul University.

Schiafﬁno, S., Garcia, P., and Amandi, A. (2008). eteacher:

Providing personalized assistance to e-learning stu-

dents. Computers & Education, 51(4):1744–1754.

Sprent, P. and Smeeton, N. C. (2016). Applied nonparamet-

ric statistical methods. CRC Press.

Study Amanuensis (November 2, 2016). private communi-

cation.

Tang, T. Y. and McCalla, G. (2005). Smart recommenda-

tion for an evolving e-learning system: Architecture

and experiment. International Journal on elearning,

4(1):105.

Tuttle, K. N. (2000). Academic advising. New directions

for higher education, 2000(111):15–24.

Wallden, L. J. (2016). Kansainv¨alisten koulutusarvioiden

vertailu koulutuksellisen tiedonlouhinnan keinoin.

Jyv¨askyl¨a Studies in Computing. University of

Jyv¨askyl¨a.

Wen, F. L. S. L. D. and McGreal, F. Z. K. R. (2015). e-

Advisor: A multi-agent system for academic advising.

International Journal of Advanced Computer Science

and Applications, 6(8).

Werghi, N. and Kamoun, F. K. (2009). A decision-

tree-based system for student academic advising and

planning in information systems programmes. In-

ternational Journal of Business Information Systems,

5(1):1–18.

Zaki, M. J. and Meira Jr, W. (2014). Data Mining and Anal-

ysis: Fundamental Concepts and Algorithms. Cam-

bridge University Press.

CSEDU 2017 - 9th International Conference on Computer Supported Education