PosterPDF Available

Mining Frequent Learning Pathways from a Large Educational Dataset



Content may be subject to copyright.
arXiv:1705.11125v3 [cs.CY] 8 Jul 2017
Mining Frequent Learning Pathways from a Large
Educational Dataset
Nirmal Patel
Playpower Labs
Collin Sellman
Arizona State University
Derek Lomas
Playpower Labs
In this paper, we describe data mining techniques used to
extract frequent learning pathways from a large educational
dataset. These pathways were extracted as a directed graph
that encoded student learning processes. Our dataset con-
tains more than 800 million interactions of over 3 million
anonymized students in an online learning platform. Per-
forming process mining on large and complex datasets reg-
ularly yields incomprehensible process models. Although,
if we cluster data and obtain groups following similar pro-
cesses, we can greatly improve process mining results. To
this end, we developed a sequence clustering algorithm that
let us group students who followed similar learning path-
ways. To extract frequent learning pathways from these
clusters of data, we developed a graph-based process dis-
covery algorithm that revealed to us the sequences of learn-
ing activities that many students followed. These sequences
represented highways of student learning.
sequence mining, process mining, graph mining, learning
Digital learning platforms collect a wide variety of student
interaction data. These data can be used to inform contin-
uous improvement at the level of the product, school dis-
trict, classroom or the individual student. Recently, we’ve
begun to investigate individual sequences of instructional
content, which we call learning pathways” or student jour-
neys.” Using various data mining techniques, we’ve analyzed
the learning pathways from thousands of students, revealing
the emergence of interesting patterns and structures.
Association rule mining has been used to find interesting
patterns within student learning logs [2]. For example, we
can find association rules like If students fail in quiz X,
what instructional content do they access afterward? Of-
ten times, we find many interesting rules, but these rules
are not linked and they do not tell us anything about entire
learning pathways of students. By using newly developed
process mining techniques, we can look for process mod-
els of student learning [6, 7]. Such models can be used to
discover usage patterns or to compare usage to ideal pat-
terns. Unfortunately, when working with data from many
thousands of students, we find that there are too many un-
derlying processes. Therefore, the resulting process models
show ”spaghetti”: an unintelligible mess of various processes.
Process discovery algorithms like Fuzzy Mining can help re-
duce the spaghetti [3], but they render models that are not
amenable to extensive manipulation or enhancement.
To address these issues, we first explored sequence clustering
techniques to see if they could help us identify groups of
students who follow similar processes [4]. We developed an
edit distance based clustering algorithm, which allowed us to
group together students with similar learning pathways. We
found sequence clustering crucial for further analysis, as it
significantly reduced the complexity of our process models.
Using our clustering technique, we discovered three distinct
groups of students that differed in their length of learning
To discover process models within the clusters, we imple-
mented a graph-based process discovery algorithm. This
algorithm uses an iterative procedure to find out how fre-
quently students transition from one activity to another in
their learning paths. Applying this algorithm to our data
yielded a directed graph. After visualizing the graph and
removing infrequent edges from it, we found an intelligible
process model. Further pruning of the graph showed us very
clearly what learning pathways students followed. We found
graph visualization to be a good tool for exploring student
activity sequences. It allowed us to quickly evaluate our
hypotheses about how students interacted with the learn-
ing platform. We also noted that our graph-based process
models were easy to manipulate.
The rest of the paper is structured as follows. In section 2,
we describe the dataset which we used for data mining. In
section 3, we briefly describe the sequence clustering algo-
rithm we used on data and discuss the results. In section
4, we succinctly describe the graph mining algorithm and
show some examples from our analysis. At the end, we dis-
cuss the usefulness of techniques presented in this paper,
and directions for future work.
Table 1: Variables of the dataset
Organizational Instructional Other
Teacher ID Activity ID Student ID
Class ID Activity Type Session ID
School ID Event Type Timestamp
District ID Skill
State Score
Our original dataset is formatted as an event log. Each row
of the dataset corresponds to an event that captures learner
interaction with an online learning platform. Examples of
events are logging in, opening an activity, saving progress in
an activity, submitting an activity etc. Millions of students
across the United States interact with the platform. Within
the platform, there are many programs available for differ-
ent subjects and grade levels. We chose to analyze a grade 3
math program. This program is a digital curriculum divided
into topics, lessons, and activities, and can be accessed via a
web browser. There are different types of activities that stu-
dents can do, including games, formative assessments, and
summative assessments. Students can either do activities
assigned by the teacher or choose to do activities they like.
Teachers can also create their own assessments and upload
them on the platform. For our analysis, we selected a par-
ticular formative assessment activity and randomly sampled
20,000 users from the population of students who access it.
We chose sample size so as to keep our computations effi-
The dataset contains many different variables, most of which
are listed in Table 1. Roughly, these variables can be put
into two major categories: organizational and instructional.
Organizational variables tell us about the student’s learning
context and instructional variables tell about the student’s
interaction with digital assets and the platform. Since we
were first interested in finding out what sequences of things
students follow, we concentrated on instructional variables.
For the formative assessment that we selected for analysis,
multiple event types are possible. We selected the submit-
ting event, which indicated that student had finished the
assessment. Typically, this event also contains the student’s
score. We transformed the data and made them amenable
to analysis by filtering the dataset and selecting particular
variables. Specifically, we chose to extract three variables
from our database: Student ID,Activity ID and Times-
tamp. This way, we had access to the sequence of activities
(in our case, formative assessments) that every student went
through. These activities spanned across the entire grade 3
math curriculum. After we had extracted data in the above
format and sampled the desired number of students, we used
a sequence clustering algorithm to group together students
with similar activity sequences.
Exploration of our large dataset revealed to us that students
rarely followed same learning paths. This fact posed a big
challenge to our process mining work since high variance in
student activity sequences kept giving us ”spaghetti” process
models. We hypothesized that by clustering together stu-
dents who followed similar learning paths, we could get more
Table 2: Clustering results
Cluster 1 Cluster 2 Cluster 3
Avg. sequence length 2.80 13.72 44.85
SD sequence length 1.91 5.79 18.91
Number of students 10524 5524 3952
intelligible process models. In the process mining paradigm,
various algorithms have been proposed to cluster sequential
data [1]. Algorithms specific to clustering student activity
sequences have also been proposed [5]. We developed and
used an edit distance based clustering algorithm to cluster
students with similar learning paths.
Edit distance (or more specifically, Levenshtein distance) be-
tween two strings is measured as the minimum number of
characters we have to add, subtract and substitute to turn
one string into another. We extended this idea to arbitrary
alphabets, so the distance between two activity sequences
was measured as the number of activities we had to add,
subtract and substitute to turn one activity sequence into
another. For example, distance between activity sequences
<a,b,b,c,d,e,e>and <a,b,c,d,e>is 2, because by removing
two activities from first sequence, we can turn it into sec-
ond sequence in the least number of operations. We used
this generalized distance metric to compute a distance ma-
trix having pairwise distances between all student activity
sequences. Several clustering methods are available to clus-
ter distance matrices. Partitioning around medoids offers a
partitioning approach but requires the number of clusters
in advance. On the other hand, hierarchical clustering does
not require the number of clusters in advance, so we used
it to cluster our distance matrix. We used Ward’s method
for agglomeration. Hierarchical clustering can be done us-
ing various agglomeration criteria. We found that using
Ward’s method as a criterion gave us clusters where obser-
vations within clusters had similar paths and path lengths.
Although other methods like average linkage could give clus-
ters with higher within cluster path similarity, they suffered
from an excessive amount of nesting. In contrast, Ward’s
method was able to produce more equally sized clusters.
Our results indicated the presence of 3 clusters which had
meaningful differences. Properties of these clusters are listed
in Table 2. Looking at the results, we inferred that Clus-
ter 1 corresponded to students who did just a few items
and stopped. Subsequent clusters appeared to be related
to students with medium and high amounts of usage in the
program. Cluster 3 students generally did more activities
than cluster 2 students, i.e. they went further in using their
digital curriculum.
To mine a graph that encoded student learning pathways,
we followed a simple procedure. We started by creating
an n×nzero matrix Mwhere nis the number of unique
activities that students did. Entry Mij corresponded to
how many times students did activities iand jin sequence.
We iterated through data of all students, and for every pair
of subsequent activities that students did, we added 1 to
the appropriate cell of M. At the end of the procedure, we
produced an adjacency matrix of a directed graph.
Table 3: Graph mining results
Cluster 1 Cluster 2 Cluster 3
Nodes 123 169 474
Edges 3277 5378 6662
(a) Unfiltered graph (b) Filtered graph
Figure 1: Learning pathway graphs
Every node in this graph corresponded to an activity that a
student could do, and edges of the graph showed pathways
that students had taken. Edge weights of the graph repre-
sented how many times that edge was traveled by students.
We ran the graph mining algorithm on clusters we had dis-
covered earlier. Table 3 compares the number of nodes and
edges of graphs corresponding to clusters.
We noted that even though the number of students de-
creased from Cluster 1 to 3, the number of nodes and edges
increased. We chose to focus on Cluster 3, where students
learning pathways were substantially longer than in other
clusters and the graph was denser. Figure 1 (a) shows the
learning pathway graph of Cluster 3, without any edge filter-
ing and changes in node size. We can see that it is impossible
to make any inferences visually (it looks like ”spaghetti”).
Even after clustering, infrequent paths visually obscured
mainstream student behavior. To see frequent paths, we
removed infrequent edges in the graph. Figure 1 (b) shows
a graph where edges with less weight are filtered out and
node sizes are proportional to node degrees. Now we see
highways of student learning, i.e. paths that many students
took. When we filtered the graph in Figure 1 (b) further, we
saw even fewer paths, and these paths could be investigated
directly by looking at the visualization. Figure 2 shows those
few paths. Figure 3 zooms into a mid left region of Figure
2 to show three activities whose paths are frequently tra-
versed. We confirmed that these activities follow each other
in the digital curriculum too.
By performing sequence clustering and graph-based process
mining on educational data, we were able to identify student
groups with similar usage patterns and examine their learn-
ing pathways. While we applied our technique to a smaller
subset of data, we believe that it is a generalizable tech-
nique that can be applied to any educational dataset from
which we can extract student activity sequences. We can
analyze student learning pathway graphs to explore how dif-
ferent paths affect student and class performance. Teachers
and curriculum designers can learn the sequence of activi-
ties their students generally go through during their learning
and whether these sequence of activities conform to a de-
Figure 2: Some highways of student learning
Figure 3: Three activities that follow each other many times
sired or designed sequence. Last but not least, designers of
digital learning products can get insights into user behavior
that can help them make informed design decisions towards
data-driven continuous improvement. We believe the meth-
ods presented in this paper can provide information that can
be of use to various stakeholders involved in education, by
showing them what students do and creating more accurate
models of student behavior in digital environments. Open
source Rimplementation of sequence clustering technique
presented in this paper is available online1.
Additional student data related to their activity and per-
formance (such as scores) can easily be added to student
pathways. We are currently mining decision points in graphs
that will explore how performance on one activity influences
the choice of the next activity which is a rich area for fu-
ture research. This is similar to decision mining in process
models [8]. Since we have skills mapped to each activity
in our dataset, we also hope to explore the skill acquisition
processes of students. In the methods domain, computa-
tional challenges related to the scalability of our procedures
also remain when using larger, unsampled datasets. We be-
lieve these can be addressed by using a parallel computing
framework like Apache Spark.
[1] R. J. C. Bose and W. M. van der Aalst. Context aware
trace clustering: Towards improving process mining
results. In Proceedings of the 2009 SIAM International
Conference on Data Mining, pages 401–412, 2009.
[2] E. Garc´ıa, C. Romero, S. Ventura, C. de Castro, and
T. Calders. Association rule mining in learning
management systems. Handbook of educational data
mining, pages 93–106, 2011.
[3] C. W. G¨
unther and W. M. Van Der Aalst. Fuzzy
mining–adaptive process simplification based on
multi-perspective metrics. In International Conference
on Business Process Management, pages 328–343.
Springer, 2007.
[4] B. Hompes, J. Buijs, W. van der Aalst, P. Dixit, and
J. Buurman. Discovering deviating cases and process
variants using trace clustering. In Proceedings of the
27th Benelux Conference on Artificial Intelligence,
pages 5–6, 2015.
[5] S. Klingler, T. K¨
aser, B. Solenthaler, and M. Gross.
Temporally coherent clustering of student data. In
Proceedings of the 9th International Conference on
Educational Data Mining, International Educational
Data Mining Society, pages 102–109, 2016.
[6] P. Mukala, J. Buijs, and W. Van der Aalst. Exploring
students’ learning behaviour in moocs using process
mining techniques. Technical report,,
[7] N. Trcka, M. Pechenizkiy, and W. Van der Aalst.
Process mining from educational data. Handbook of
Educational Data Mining, pages 123–142, 2011.
[8] W. M. van der Aalst. Process mining: data science in
action. Springer, 2016. pg. 294.
... Reference Graphical behavior models [Okamoto et al. 2014], [Brown et al. 2015], [Kumar 2014], [Jiang et al. 2014] Graph-based log analysis [Gitinabard et al. 2017], [Dawson 2008], [Gardner and Brooks 2018], [Grawemeyer et al. 2017], [Kovanovic et al. 2014], [Mostafavi and Barnes 2014], Graphical solution representations , [Li et al. 2015], , [Gašević et al. 2013], [Eckles and Stradley 2012], [Fire et al. 2012], [Dekel and Gal 2014] Novel graph analysis techniques [Buluz and Yilmaz 2017], [Patel et al. 2017], [Costa et al. 2019], [London and Németh 2014], [Zheng et al. 2017] Tools and technologies for automatic concept hierarchy extraction [Chen et al. 2018], [Cateté et al. 2014], [Yang et al. 2010] Tools and technologies for traph grammar (pattern) recognition [Lynch 2014] Relevant analytical tools and standard problems [Sheshadri et al. 2014] Social network analysis [Wejnert 2010], [Kovanovic et al. 2014], [Garcia-Saiz et al. 2014] The Directed Acyclic Graph is the most widely used representation (34%), followed by Directed Acyclic Multigraph with 23% (see Table 2). In general Directed Acyclic Graph are graphs are often used in analysis of user interactions in virtual environments. ...
... Directed Acyclic Graph [Buluz and Yilmaz 2017], [Chen et al. 2018], [Kumar 2014], [Cateté et al. 2014], [Costa et al. 2019], [Wejnert 2010], [London and Németh 2014], [Zheng et al. 2017], , [Gitinabard et al. 2017] Directed Cyclic Graph [Yang et al. 2010] Directed Acyclic Multigraph [Dawson 2008], , [Li et al. 2015], [Brown et al. 2015], [Patel et al. 2017], [Kovanovic et al. 2014], [Ramos et al. 2016] Simple Acyclic Graph , [Gašević et al. 2013], [Grawemeyer et al. 2017], [Lynch 2014] , [Jiang et al. 2014] Tree [Gardner and Brooks 2018], [Dekel and Gal 2014], [Mostafavi and Barnes 2014], [Sheshadri et al. 2014 [Zheng et al. 2017] [London and Németh 2014] [Dekel and Gal 2014]. One focused on grammar models for learning [Lynch 2014] [Kumar 2014] and the others on tools ] [Ramos et al. 2016] [Chen et al. 2018], process learning and Social Network analysis [Fire et al. 2012] [Yang et al. 2010] ] [Gardner and Brooks 2018] [Eckles and Stradley 2012] [Wejnert 2010]. ...
... Python NetworkX [Grawemeyer et al. 2017] [Fire et al. 2012] ] is a Python script for graph analysis and visualization, and offers one of the largest varieties of algorithms for graph analysis. The Rgraph plugin for R [Fire et al. 2012], [Patel et al. 2017] presents some algorithms and metrics for visualization and extraction of information in graphs. ...
... By examining students' learning behavior patterns, we can assess the curriculum and evaluate students' performance during the study period. Some works on EPM have been applied in academic institutions to predict students' dropout [2], to recommend a correct path to students [3], to identify, examine and facilitate the evaluation of learning processes [4]- [6], and to enhance the curriculum design [3], [7], [8]. ...
... The curriculum guideline is designed for eight (8) consecutive semesters, which is the standard length of study for bachelor's degrees. During the eight (8) semesters, there are two phases, i.e., the preparation phase (1st semester to 4th semester) and the bachelor's phase (5th semester to 8th semester). ...
Full-text available
Educational process mining is one of the research domains that utilizes students’ learning behavior to match students’ actual courses taken and the designed curriculum. While most works attempt to deal with the case perspective (i.e., traces of the cases), the temporal case perspective has not been discussed. The temporal case perspective aims to understand the temporal patterns of cases (e.g., students’ learning behavior in a semester). This study proposes an extension of cluster evolution analysis, called profile-based cluster evolution analysis, for students’ learning behavior based on profiles. The results show three salient features: (1) cluster generation; (2) within-cluster generation; and (3) time-based between-cluster generation. The cluster evolution phase extends the existing cluster evolution analysis with a dynamic profiler. The model was tested on actual educational data of the Information System Department in Indonesia. The results showed the learning behavior of students who graduated on time, the learning behavior of students who graduated late, and the learning behavior of students who dropped out. Students changed their learning behavior by observing the migration of students from cluster to cluster for each semester. Furthermore, there were distinct learning behavior migration patterns for each category of students based on their performance. The migration pattern can suggest to academic stakeholders students who are likely to drop out, graduate on time or graduate late. These results can be used as recommendations to academic stakeholders for curriculum assessment and development and dropout prevention.
... Patel et al. [8] analyses the LPs commonly used by students. The analysis uses sequence clustering and graph-based process mining on educational data. ...
This paper promotes the idea of the learning process management in the e-learning system. A personalized adaptive e-learning system is used in this research that comprises three developed topic acquisition sequences: teacher, learner or optimal topic sequences. The learner has the ability to switch between the aforementioned topic sequences. The system stores data about the course acquisition process. The analysis of the stored data demonstrated that a bit more than half of the students used the teacher topic sequence; higher grades in topics got those students who chose the learner or optimal topic sequence; the grades of the half of the students who used the optimal and teacher topic sequences were in the same level. The obtained results were used as the justification for the improvement of the existing optimal topic sequence development method. As a result, an algorithm for the recommended learning path development is proposed in this paper. The topics of the course and links in between are described using a weighted directed graph. The weight of every edge and vertex of the graph is calculated based on the parameter values describing the topic. Afterwards, the recommended learning path is assumed to be the path with the lowest weight that is found in the weighted oriented graph using a search.
... In the paper [11] Patel N. et al. describe data mining techniques used to extract frequent learning pathways from a large educational dataset. A graph-based process discovery algorithm was developed for extracting these frequent learning pathways from the clusters of data and revealed the sequences of learning activities that many students followed. ...
Conference Paper
Full-text available
Abstract The article presents a graph model for building learning pathways between academic courses based on the relations between keywords. In this study, we use two-layer graph model where the first layer is represented by academic courses, the second one by the keywords. Relations between courses are built based on the relations between the keywords. An approach to individual learning pathway validation by the syllabus keywords graph is proposed.
Full-text available
A detailed illustration of how large scale digital learning systems can incrementally reduce the poverty-achievement gap
Full-text available
Learning to read is one of the most important achievements of early childhood, and sets the stage for future success. Even prior to school entry, children’s foundational literacy skills predict their later academic trajectories (Duncan et al., 2007; La Paro & Pianta, 2000; Lloyd, 1969; Lloyd, 1978). Children learn to read with differing levels of ease, with an estimated 5-17% of school-age children who struggle with reading acquisition (Shaywitz, 1998). The individual variation in children’s reading skills can be predicted by genetic, environmental, academic and socio-demographic factors (for review, see Peterson & Pennington, 2015). This chapter focuses on the relationship between reading development and socioeconomic status (SES), with attention to both cognitive outcomes and neural mechanisms. First, we describe SES and its relation to academic achievement in general, and reading development in particular. Second, we examine environmental factors that can potentially give rise to socioeconomic disparities in reading, such as early language/literacy exposure and access to books. Next, we explore the link between SES and reading disability (RD), including a focus on intervention approaches and treatment response. Finally, we summarize remaining questions and propose future research priorities.
Full-text available
Durante las últimas dos décadas, la investigación en el área de estudio de la pobreza ha comenzado a aportar evidencia que constituye un avance en la comprensión de cómo la adversidad temprana asociada a privaciones materiales, sociales y culturales modula el desarrollo cerebral. Cuando tal evidencia es utiliada en otros contextos disciplinares, con frecuencia se verifican referencias al desarrollo cerebral temprano como a un factor predictor de conductas adaptativas y de productividad económica durante la vida adulta; o de la imposibilidad de tales logros por la supuesta inmutabilidad de los impactos negativos a largo plazo de la pobreza infantil. Este tipo de afirmaciones, que tienen implicaciones no solo científicas sino también políticas, requieren ser analizadas de manera adecuada a la luz de la evidencia disponible, ya que podrían inducir conceptos erróneos y sobre-eneralizaciones que tienen la potencialidad de afectar los criterios para la inversion, el diseño, la implementación y la evaluación de acciones en el ámbito de la infancia temprana. Este libro se propone como un aporte en esta dirección. Los distintos capítulos, a cargo de referentes destacados del estudio neurocientífico y cognitivo de la pobreza, aportan evidencia que alimenta hipótesis y reflexiones en línea con las principales preguntas de tal área de estudio.
Conference Paper
Full-text available
Information systems supporting business processes generate event data which provide the starting point for a range of process mining techniques. Lion's share of real-life processes are complex and ad-hoc, which creates problems for traditional process mining techniques, that cannot deal with such unstructured processes. Finding mainstream and deviating cases in such data is problematic, since most cases are unique and therefore determining what is normal or exceptional may depend on many factors. Trace clustering aims to group similar cases in order to find variations of the process and to gain novel insights into the process at hand. However, few trace clustering techniques take the context of the process into account and focus on the control-flow perspective only. Outlier detection techniques provide only a binary distinction between normal and exceptional behavior, or depend on a normative process model to be present. As a result, existing techniques are less suited for processes with a high degree of variability. In this paper, we introduce a novel trace clustering technique that is able to find process variants as well as deviating behavior based on a set of selected perspectives. Evaluation on both artificial and real-life event data reveals that additional insights can consequently be achieved.
Conference Paper
Full-text available
Process Mining is a technique for extracting process models from ex- ecution logs. This is particularly useful in situations where people have an ide- alized view of reality. Real-life processes turn out to be less structured than peo- ple tend to believe. Unfortunately, traditional process mining approaches have problems dealing with unstructured processes. The discovered models are often "spaghetti-like", showing all details without distinguishing what is important and what is not. This paper proposes a new process mining approach to overcome this problem. The approach is configurable and allows for different faithfully simpli- fied views of a particular process. To do this, the concept of a roadmap is used as a metaphor. Just like different roadmaps provide suitable abstractions of reality, process models should provide meaningful abstractions of operational processes encountered in domains ranging from healthcare and logistics to web services and public administration.
This is the second edition of Wil van der Aalst’s seminal book on process mining, which now discusses the field also in the broader context of data science and big data approaches. It includes several additions and updates, e.g. on inductive mining techniques, the notion of alignments, a considerably expanded section on software tools and a completely new chapter of process mining in the large. It is self-contained, while at the same time covering the entire process-mining spectrum from process discovery to predictive analytics. After a general introduction to data science and process mining in Part I, Part II provides the basics of business process modeling and data mining necessary to understand the remainder of the book. Next, Part III focuses on process discovery as the most important process mining task, while Part IV moves beyond discovering the control flow of processes, highlighting conformance checking, and organizational and time perspectives. Part V offers a guide to successfully applying process mining in practice, including an introduction to the widely used open-source tool ProM and several commercial products. Lastly, Part VI takes a step back, reflecting on the material presented and the key open challenges. Overall, this book provides a comprehensive overview of the state of the art in process mining. It is intended for business process analysts, business consultants, process managers, graduate students, and BPM researchers.
Conference Paper
The extraction of student behavior is an important task in educational data mining. A common approach to detect similar behavior patterns is to cluster sequential data. Standard approaches identify clusters at each time step separately and typically show low performance for data that inherently suffer from noise, resulting in temporally inconsistent clusters. We propose an evolutionary clustering pipeline that can be applied to learning data, aiming at improving cluster stability over multiple training sessions in the presence of noise. Our model selection is designed such that relevant cluster evolution effects can be captured. The pipeline can be used as a black box for any intelligent tutoring system (ITS). We show that our method outperforms previous work regarding clustering performance and stability on synthetic data. Using log data from two ITS, we demonstrate that the proposed pipeline is able to detect interesting student behavior and properties of learning environments.
Learning management systems(LMs) can offer a great variety of channels and workspaces to facilite information haring and communication aming participants in a course. The let educators distribute informaction to students, produce content material, prepare assignments and tests, engage in discussions, manage distance classes, and enable collaborative learning with forums, chats, file storage areas, news services... Some examples of commercial systems are Blackboard[1], WebCT[2], and Top-Class[3], while some examples of free systems are Moodle[4], Ilias[5], and Claroline[6]. One of the most commonly used is Moodle (modular object oriented developmental learning environment), a free learning management system enabling the creation of powerful, flexible, and engaging online courses and experiences[42]. These e-learning systems accumulate a vast amount of information that is very valuable for analyzing students´ behavior and could create a gold mine of educational data[7]. They can record any student activities involved, such as reading, writing, taking tests, performing various tasks, and even communicating with peers. They normally also provide a data base that stores all the system´s information: personal information about the users (profile), and academic results and users´ interaction data. However, due to the vast quantities of dat these systmes can generate daily, it is very difficult to manage data analysis manually.
In modern education, various information systems are used to support educational processes. In the majority of cases, these systems have logging capabilities to audit and monitor the processes they support. At the level of a university, administrative information systems collect information about students, their enrollment in particular programs and courses, and performance like examination grades. In addition, the information about the lectures, instructors, study programs, courses, and prerequisites are typically available as well. These data can be analyzed from various levels and perspectives, showing different aspects of organization, and giving us more insight into the overall educational system. From the level of an individual course, we can consider participation in lectures, accomplishing assignments, and enrolling in midterm and final exams. However, with the development and increasing popularity of blended learning and e-learning, information systems enable us to capture activities also at different levels of granularity. Besides more traditional tasks like overall student performance or dropout prediction [1], it becomes possible to track how different learning resources (videolectures, handouts, wikis, hypermedia, quizzes) are used [2], how students progress with (software) project assignments (svn commits) [3], and self-assessment test and questionnaires [4].
Exploring students' learning behaviour in moocs using process mining techniques
  • P Mukala
  • J Buijs
  • W Van Der Aalst
P. Mukala, J. Buijs, and W. Van der Aalst. Exploring students' learning behaviour in moocs using process mining techniques. Technical report,, 2015.