PreprintPDF Available

Abstract and Figures

This paper presents an approach of using methods of process mining and rule-based artificial intelligence to analyze and understand study paths of students based on campus management system data and study program models. Process mining techniques are used to characterize successful study paths, as well as to detect and visualize deviations from expected plans. These insights are combined with recommendations and requirements of the corresponding study programs extracted from examination regulations. Here, event calculus and answer set programming are used to provide models of the study programs which support planning and conformance checking while providing feedback on possible study plan violations. In its combination, process mining and rule-based artificial intelligence are used to support study planning and monitoring by deriving rules and recommendations for guiding students to more suitable study paths with higher success rates. Two applications will be implemented, one for students and one for study program designers.
Content may be subject to copyright.
A Combined Approach of Process Mining and
Rule-based AI for Study Planning and
Monitoring in Higher Education
Miriam Wagner1, Hayyan Helal2, Rene Roepke3, Sven Judel3, Jens Doveren3,
Sergej Goerzen3, Pouya Soudmand1, Gerhard Lakemeyer2, Ulrik Schroeder3,
and Wil van der Aalst1
1Process and Data Science (PADS), RWTH Aachen University, Germany
2Knowledge-Based Systems Group, RWTH Aachen University, Germany
3Learning Technologies Research Group, RWTH Aachen University, Germany
Abstract. This paper presents an approach of using methods of process
mining and rule-based artificial intelligence to analyze and understand
study paths of students based on campus management system data and
study program models. Process mining techniques are used to character-
ize successful study paths, as well as to detect and visualize deviations
from expected plans. These insights are combined with recommendations
and requirements of the corresponding study programs extracted from
examination regulations. Here, event calculus and answer set program-
ming are used to provide models of the study programs which support
planning and conformance checking while providing feedback on possible
study plan violations. In its combination, process mining and rule-based
artificial intelligence are used to support study planning and monitor-
ing by deriving rules and recommendations for guiding students to more
suitable study paths with higher success rates. Two applications will be
implemented, one for students and one for study program designers.
Keywords: Educational Process Mining ·Conformance Checking ·Rule-
based AI ·Study Planning ·Study Monitoring.
1 Introduction
In higher education, study programs usually come with an idealized, recom-
mended study plan. However, given how students have different capacities to
study due to circumstances like part-time jobs or child care, and how one de-
viation from the intended study plan can have ripple effects spanning several
semesters, in reality, a large number of different study paths can be observed.
Further, capacity limits like room sizes or the amount of supervision that lectur-
ers can provide make the planning of study paths more complex. Even though
individualized study paths are possible due to the flexibility in study programs
arXiv:2211.12190v1 [cs.AI] 22 Nov 2022
2 M. Wagner et al.
and their curriculum, students may need assistance and guidance in planning
their studies. Software systems that assist students and study program design-
ers in planning might do so by analyzing the large amounts of data in higher
education institutions [12]. Of particular interest in this context are event data
extracted from Campus Management Systems (CMS) including course enroll-
ments, exam registrations and grade entries. To this purpose the AIStudyBuddy
project - a cooperation between RWTH Aachen University, Ruhr University
Bochum and University of Wuppertal - is set up. For preliminary analyses, we
received access to the CMS data of two Bachelor programs, Computer Science
and Mechanical Engineering, at RWTH Aachen University. Within the project,
it will be investigated how to preprocess the data of all partners to apply the
preliminary as well as the further intended analyses.
Fig. 1. Overview of the project, showing the two parts: StudyBuddy and BuddyAna-
lytics and their relationships to the different systems and techniques.
The aim of the project is to develop two applications: an intelligent planning
tool for students and an analytics dashboard for study program designers (see
Figure 1). Both will be powered by a combination of rule-based Artificial In-
telligence (AI) and Process Mining (PM) approaches. The implementation and
evaluation of this combination’s ability to efficiently generate rich feedback when
checking the conformance to formal study plans is a key aspect of this project.
This feedback will include PM results in the form of recommendations, which do
not result from explicit regulations but rather historic study path data.
The planning tool for students, StudyBuddy, will use rule-based AI to check
preliminary plans against an encoding of study program regulations. It will be
Process Mining and Rule-based AI for Study Planning and Monitoring 3
able to provide immediate high-quality feedback regarding any potential con-
flicts in students’ study plans. In addition to the rules explicitly codified in
institutional regulations, the tool will have a notion of recommendations, which
result from analyzing historical CMS data using PM approaches and finding
characterizations of successful paths, e.g., finished in standard period of study.
The analytics dashboard, BuddyAnalytics, will enable study program design-
ers to explore the PM results for the process of Curriculum Analytics. Process
models of recommended study plans can be compared to study paths in the data
to detect possible deviations or favorable routes. Various study path analyses
could support monitoring and help study program designers as well as student
counseling services to support successful study paths and intervene in misguided
study planning by providing individualized plans.
The paper is structured as follows: Section 2 presents relevant related work
in the fields of PM, rule-based AI and curriculum analytics. Section 3 introduces
the aim of addressing individualized study planning for students and data-driven
study monitoring for study program designers in a combined approach. The
current state as well as challenges of the project are described in Section 4, while
Section 5 presents objectives of future work. Section 6 concludes the paper.
2 Related Work
2.1 Process Mining in Education
Educational Process Mining (EPM) [4,27] is a sub-field of PM [28], using var-
ious, commonly known PM techniques in the educational context, e.g. higher
education. While we focus on CMS data, most work in EPM has been done
using Learning Management Systems (LMS) data with similar aims. In [20],
two online exams have been analyzed using dotted chart analysis and process
discovery with various miners. The applicability of standard methods provided
in ProM in the context of LMS data is shown. In [5], course-related student
data has been extracted to visualize the learning processes using an inductive
miner to help preventing failing the course. “Uncover relationships between us-
age behavior and students’ grades” is the aim of [13] by using Directly-Follow
Graph (DFG). In [11], a case study is described in which the LMS data of one
course is analyzed using among other things DFG. Also, in [18], data from an
LMS is used and the creation of the event log is described in detail. Those event
logs are used for the creation of DFG.
Analyses of LMS data show that the PM techniques can be used in the
educational context but while concentrating on the behavior of students in one
course, Curriculum Mining analyzes the different study paths a student can
take [19] which is a substantial aspect in our work. Here, different approaches
exist: [29,25] describe ways to use curriculum data to uncover the de-facto paths
students take to, in the next step, recommend suitable follow-up courses. To our
knowledge, this next step has not been done. [8] focuses on the study counselor
perspective and uses, e.g., Fuzzy Miner and Inductive Visual Miner, to visualize
4 M. Wagner et al.
the de-facto study paths and use those insights to improve the curriculum. In [23],
the influence of failing a course on the study success is analyzed using mainly
DFGs, while in [24], the analysis is done by modeling how students retake courses
and the influence on study program dropouts.
Further, we will explore the application of conformance checking [10]. There-
fore, similar approaches to ours are reviewed. An extended approach to confor-
mance checking is multi-perspective conformance checking as in [17]. For our
purpose, one reason to not extend this technique is that the Petri nets repre-
senting the study plan are hard to read when including all allowed behavior. For
example, allowing a course to be in different semesters might lead to reposition-
ing other courses as well. Furthermore, some rules that need to be represented
are not connected to the model itself, e.g., credit point thresholds belonging to a
semester and not to a course. Those could be modeled using invisible transitions,
which makes the model more complicated and less intuitive.
2.2 Related Work on Rule-based AI
The goal of rule-based AI is to model the examination regulations and the module
catalog in a machine-readable language that allows for dealing with and planning
events. For such scenarios, the combination of Answer Set Programming (ASP)
and Event Calculus (EC) is applied. Both are based on a wider concept called
non-monotonic reasoning, which differentiates from monotonic reasoning by the
ability to retract already made implications based on further evidence [6].
Non-monotonic reasoning can model defaults as described in [22]. Defaults
are assumed to hold, but do not necessarily have to. For instance, Students
typically take course X after they do course Y will be modeled as a default,
as it is a recommendation, not a requirement. As long as the student does not
plan anything against it, it will be considered in their planning. Else, it will be
ignored. A requirement on the other hand must be valid for all plans.
Looking for similar approaches, in [2], the problem of curriculum-based course
timetabling was solved using ASP, however using a mechanism other than EC.
While we consider recommendations to be defaults that must be typically fol-
lowed, they should only ever result in a warning to the student, still giving the
freedom to be deviated from. In [2], recommendations come in handy for plan-
ning, where the number of violations on them should be minimized. Furthermore,
the timetabling problem focuses much more on the availability requirement for
courses rather than also considering the results (e.g. success or failure, Credit
Points (CPs) collected, ...) of these courses, which is the main focal point for us.
More generally, Declarative Conformance Checking [10] is a common appli-
cation of rule-based AI to process models. In [16,9], declarative rules are used in-
stead of classical conformance checking based on Petri nets. While [16] just covers
the activities for constraints, [9] extended it with a data- and time-perspective.
Furthermore, [1] has a wider model for requirements. It specifies three kinds of
requirements, which refer to the relation in time between events, e.g. an event
has a succession requirement if there is an event that must be done in the future
after doing it. All three approaches use Linear Temporal Logic instead of ASP
Process Mining and Rule-based AI for Study Planning and Monitoring 5
and EC, as it it suitable for modeling the three mentioned requirements. For
our purposes though, it makes the modeling of the contribution of an event to
a specific result (e.g., CPs) harder, because our approach does not focus on the
relation in time between events as much as the contributions of these events.
2.3 Curriculum Analytics and Planning
Having emerged as a sub-field of Learning Analytics, curriculum analytics aims
to use educational data to drive evidence-based curriculum design and study
program improvements [15]. Leveraging the data gathered in educational in-
stitutions, it can help identify student’s needs and reduce dropout rates [12].
As such, different approaches and tools (e.g., [3,7,14,21]) have been developed
to support the analysis of CMS or LMS data with the aim of helping instruc-
tors and program coordinators reflect on the curriculum and teaching practices.
While various data and PM approaches have been used to analyze study paths
provided through CMS event data [3,21], curriculum sequencing and study plan-
ning was explored using semantic web concepts applied on examination regula-
tions, with the overall aim of supporting curriculum authoring, i.e., the design
of personalized curricula fulfilling a set of constraints [1]. Other approaches in-
clude recommender systems [30] or genetic algorithms [26] to support students
in course selection processes and fulfilling requirements of a study program. To
the best of our knowledge, however, no joint approach of PM and rule-based AI
has yet been explored in order to support study planning and monitoring for
students and study program designers.
3 Approach
The aim of AIStudyBuddy is to support individualized study planning (for stu-
dents) and monitoring (for study program designers). Study planning describes
the students’ activities of planning and scheduling modules, courses and exams
throughout the complete course of a study program. The examination regula-
tions provide recommendations and requirements to describe a study program
and the conditions for students to successfully earn a degree. These may include
a sample study plan recommending when to take which module or course and
attempting to distribute CPs evenly over the standard period of study. Students
choose from the module catalog, a list of mandatory and elective modules.
While most students may start with the same recommended plan in their first
semesters, deviations due to various reasons can occur at any time, e.g., working
part-time may result in a reduced course load and delaying courses to the next
year, thus, changing the complete plan and its duration. Therefore, support for
individualized planning as well as recommendations of suitable study paths are
needed. Further, the diversity of study paths and deviations from recommended
study plans raises questions of how different students move through a study
program, if certain modules or courses cause delays in the study plan, or whether
a study program may need revisions. Here, study monitoring can be provided
6 M. Wagner et al.
by analyzing students’ traces in various systems used in the university. In our
project, we will initially focus on CMS data and might include LMS data later.
In order to support students and study program designers in their respective
tasks, a modular infrastructure (see Figure 1) with two primary applications for
the target groups will be implemented. The application StudyBuddy presents a
web interface to guide and engage students in study planning activities. As in
many programs students do not necessarily have to follow a recommended plan
and in later phases not even have recommendations. To help finding suitable
courses historic data can be used to give hints which combinations have been
successful. Furthermore, course-content is not always independent from other
courses and a specific order might help to pass with higher chance. It offers an
overview of a student’s study program and allows for creation and validation of
individual study plans. ASP and EC are used to model these regulations. Given
a study plan, they can be used to generate feedback regarding violations and
give recommendations. These recommendations are the result of mining historic
data of previous study paths for those with high success rates.
For study program designers, the application BuddyAnalytics presents an in-
teractive, web-based dashboard visualizing PM data analysis results. Different
methods, i.e., process discovery and conformance checking, can help to under-
stand how different student cohorts behave throughout the course of the study
program and identify deviations from recommended study plans. Based on dif-
ferent indicators and questions by study program designers, student cohorts can
be analyzed and insights into their paths can be gained. Study program design-
ers can evaluate and compare different study paths and further develop new
redesigned study plans in an evidence-based way.
4 Current State & Challenges
The main data source for this project is the CMS of a university, which contains
information about the students, courses, exams and their combination. Later,
the possibility to integrate LMS data will be explored. As the project aims to
be independent from the systems and study programs at different universities, a
general data model has been created (see Figure 2). This model is the starting
point for our project work and shows the general relation between courses and
students as well as study programs. The diagram does not include all possible
data fields as they differ depending on the available data of a university.
Students can have multiple study programs, e.g., first do a Bachelor in Com-
puter Science followed by a Master. Each semester a student has a study status,
e.g., enrolled or semester on leave. The same offered course is scheduled in differ-
ent semesters, e.g., Programming is offered every winter semester, and in different
study programs, e.g., Introduction to Data Science is mandatory for a Master in
Data Science but elective for a Master in Computer Science. Students also have
a status for scheduled courses during their study program, e.g., course passed.
Until now, we explored data on exam information (ie., registrations and re-
sults). The analyzed data includes Bachelor and Master Computer Science stu-
Process Mining and Rule-based AI for Study Planning and Monitoring 7
Fig. 2. A basic and generic data model for CMS data
dents as well as Mechanical Engineering Bachelor of RWTH Aachen University.
Some standard KPIs used in various departments of universities that give mean-
ingful insights about students, study programs or cohorts are:
Success rate of a course [in specific semesters] [for a cohort]
Number of attempts a course is taken [on average] [for a cohort]
Exams taken/passed in a [specific] semester [on average] [of a cohort]
Average study duration [of a cohort]
Percentage of dropouts [of a cohort] [in a predefined period]
A cohort can be defined based on the semester a group of students started,
e.g., cohort WS21 refers to all students that started in winter semester 2021/2022.
It can also be defined by the amount of semesters students already studied or the
examination regulations they belong to. Different cohort definitions are needed
to answer various questions about the behavior of students.
For more insights exceeding simple SQL queries used for descriptive statistics,
the data is transferred into specific event logs, in which activities can be based
just on courses and exams, or may even include additional information. First,
we concentrated on events describing the final status of exams for students. A
student can have multiple occurrences of a course, e.g. when they do not pass
the exam in the first try or when they registered first, but in the end, they did
not take it. As a timestamp, the semester or the exact exam date can be used.
Also, some activities may have specific status dates, e.g., the date of the (de-
)registration. Those event logs can be used to create de-facto models showing the
actual behavior of a group of students. As model we use DFG, BPMN models,
process trees and Petri nets, as shown in Figure 3, because they are easy to read
also for non-specialists in PM.
For useful insights, the multiple occurrence and the partial order of courses
must be treated. The partial order is caused by using, e.g., the scheduled semester,
instead of the arbitrarily set exam dates, based on among others room capacities.
We tried out different solutions with the setting depending on the underlying
questions that should be answered by the resulting model, e.g., when using a
8 M. Wagner et al.
Fig. 3. Model created by ProM plugin "Mine Petri net with Inductive Miner" for data
of students studying in examination regulation 2018 just using their mandatory courses
combination of exam attempt and course ID as the activity, the resulting de-
facto model shows how courses are retried and visualizes better the real workload
per semester. In Figure 3, just the first occurrence of the course is used and all
exams of a semester have the same date. Semester-blocks are visible, especially
when the offered semester of a course is in mind, e.g., Programming and Calcu-
lus are offered in the winter semester. The examination regulation of 2018 states
that it should be taken in the first semester. Compared to the (simplified) rec-
ommended plan (see Figure 4) Mentoring occurs two semesters before Calculus,
while they should be concurrent. Data Communication and Security is taken
two semesters earlier than planned and before courses that should precede it,
e.g., Computability and Complexity. Those models give a first impression of the
actual study path but need interpretation.
As a simpler approach to the later proposed combination of ASP and classical
PM conformance checking, we explored the possibility of creating de-jure models
based on the recommended study plan. We used Petri nets since they can cover
course concurrency and are still understandable by non-experts. The de-jure
model in Figure 4 shows the main recommended path. Note, the data was just
available including the third semester and later courses are invisible. Using Petri
nets and conformance checking this recommendation becomes a requirement.
The results of classical conformance checking are still useful to find typical
deviation points, e.g., Linear Algebra tends to be taken in a different semester
than proposed. Also, when filtering on the first exam attempts, the resulting
insights are different from filtering on the successful passing of exams. Filtered on
the first attempt, we can see how many students actually tried to follow the plan,
while filtered on the passed exams indicates the success route. When we have a
Process Mining and Rule-based AI for Study Planning and Monitoring 9
Fig. 4. Conformance checking result using ProM plugin "Replay a Log on Petri Net for
Conformance Analysis" on data of students studying in examination regulation 2018
and a simplified Petri net model of the regulation
high percentage of students that try to follow the recommended study plan, but
just a low percentage that succeeds, this may be a warning for study program
designers that the rules may need to be adapted to improve the recommendation
and thereby increase the study success of students.
Our findings show that in later semesters, students deviate more from the
recommended study plan, which can be explained by delays occurring earlier
in their study. What is not modeled by the Petri net here is that for Seminar
(semester 5), Proseminar (semester 2) is a prerequisite. Therefore, Proseminar
has to be taken before Seminar and forms a requirement. Including those addi-
tional requirements and all already planned exceptions from the original plan,
those models are fast becoming so called spaghetti models and lose a lot of their
readability. Lastly, additional constraints, e.g., credit point constraints such as
at the end of the third semester, at least 60 CPs must have been earned, are not
taken into account using just the described approach.
For that matter, we used the combination of ASP and EC such that e.g.
defaults can model recommendations. The first main issues concerning modeling
study requirements in general and using EC was translating examination reg-
ulations given in natural languages into formal languages. We encountered the
following problems and challenges:
There are rules that are understood by any human and thus not written.
There is a lot of human interference that allows for exceptions. Exceptions
in study plans are not rare.
There are older versions of the examination regulations, which certain stu-
dents still follow.
The second problem we encountered with EC is that almost all events con-
tribute to a single result (e.g. CPs), instead of a majority of events, each initiating
new kinds of results. EC is designed for the latter, but in study plans the former
holds. We thus adjusted the EC. One modification was to differentiate between
events that happened and events that are planned. For planning in the future,
one needs to follow the rules. For events in the past, a record is sufficient and
there is no need for further requirement checking. This allows to add exceptions
that are actually inconsistent with the examination regulations. It was also im-
portant to keep track of certain relevant numbers a student has at any point in
10 M. Wagner et al.
time, in order to be able to do requirement checking. This was achieved through
results, which events can contribute to. Mathematics 1, for example, adds 9 units
to the result credit points, after the event of success at it. A requirement on CPs
should consider the general number of CPs collected or just within a field or a
time frame. For that matter we created the notion of a result requirement, which
makes sure that the sum of results caused by a subset of events is less than,
greater than, or equal to some value. With all of this in mind, we separated the
required rules into three categories:
Invariant: Rules about the requirements and the EC modified axiom system.
Variant by Admins: Rules about modules and their availability.
Variant by Student: Rules about the plan of the student.
After that, we were able to translate the examination regulations, module
catalogs, and student event logs into rules. This enables us to perform model as
well as conformance checking.
5 Future steps
Until now, the data are limited to information about exams and is exclusively
derived from the CMS. In a next step, course enrollments will be added to further
analyze study behavior of students. This additional information will give more
concrete insights about the students’ intended study plan, since at many univer-
sities, course enrollments are not automatically coupled to exam registrations.
While students might start to take a course in the intended semester, thus enroll
in it, they might realize that the workload is too high or exam qualification re-
quirements are not fulfilled and refrain from registering for the exam in the end.
This may also be valuable information considering the instructors’ workload as
more course enrollments indicate more work during the course and may require
larger lecture halls or additional support staff. As such, this workload needs to
be balanced out when planning courses for upcoming semesters
The information stored in the LMS contains valuable information to un-
derstand students’ learning behavior, as shown in related work. When combined
with activities in the CMS, a more complete view on students’ behavior and more
direct feedback about the success of the intended plan can be provided. This feed-
back can then be used in BuddyAnalytics to help study program designers in
improving curricula and recommended study plans, as well as give more informed
suggestions for individual study plans. Possibly, in StudyBuddy, students might
be informed about their behavior deviating from a recommended plan and are
presented with suggestions suitable to their individual circumstances.
On the theoretical side, the possibilities of a combination of AI and PM are
further explored and implemented. The main focus will be to improve the con-
formance checking results. Also, PM conformance checking possibilities will be
further explored. One planned aspect is the extraction of constraints from event
logs directly. We expect to learn rules that are not intended but are beneficial,
e.g., Statistics is a good preparation for Introduction to Data Science and when
Process Mining and Rule-based AI for Study Planning and Monitoring 11
taken in order, the grade and success rate of the latter improves. Those rules
could be added to the examination regulations rules as defaults.
6 Conclusion
The AIStudyBuddy project will combine different existing AI and PM frame-
works and extend them with new features, making use of the already existing
data at universities, to help students and study program designers make more in-
formed decisions about study paths and curricula. The first results get positive
feedback from students and study program designers. Currently, only a small
fraction of available CMS data was used to produce these results, leaving a lot
of potential for future steps. PM techniques already give valuable new insights to
the study program designers and the combination of AI and PM for conformance
checking in particular helps overcome restrictions due to the data and rule prop-
erties. Having requirements and recommendations, credit point boundaries, and
long-term relations between courses should be included in the system to model
examination regulations in a more accurate manner.
Acknowledgements: The authors gratefully acknowledge the funding by the
Federal Ministry of Education and Research (BMBF) for the joint project AI-
StudyBuddy (grant no. 16DHBKI016).
Printversion This paper is a postprint version. The published version is ©Springer
(DOI pending).
1. Baldoni, M., Baroglio, C., Brunkhorst, I., Henze, N., Marengo, E., Patti, V.: Con-
straint modeling for curriculum planning and validation. Interactive Learning En-
vironments 19(1) (2011)
2. Banbara, M., Inoue, K., Kaufmann, B., Okimoto, T., Schaub, T., Soh, T., Tamura,
N., Wanko, P.: teaspoon: Solving the Curriculum-Based Course Timetabling Prob-
lems with Answer Set Programming. Annals of Operations Research 275(1) (2019)
3. Bendatu, L.Y., Yahya, B.N.: Sequence Matching Analysis for Curriculum Devel-
opment. Jurnal Teknik Industri 17(1) (2015)
4. Bogarín, A., Cerezo, R., Romero, C.: A survey on educational process mining.
WIREs: Data Mining and Knowledge Discovery 8(1) (2018)
5. Bogarín, A., Cerezo, R., Romero, C.: Discovering learning processes using inductive
miner: A case study with learning management systems (LMSs). Psicothema (2018)
6. Brewka, G., Dix, J., Konolige, K.: Nonmonotonic reasoning: an overview, vol. 73.
CSLI (1997)
7. Brown, M., DeMonbrun, R.M., Teasley, S.: Taken Together: Conceptualizing Stu-
dents’ Concurrent Course Enrollment across the Post-Secondary Curriculum Using
Temporal Analytics. Learning Analytics 5(3) (2018)
8. Buck-Emden, R., Dahmann, F.D.: Analyse von Studienverläufen mit Process-
Mining-Techniken. HMD Praxis der Wirtschaftsinformatik 55(4) (2018)
12 M. Wagner et al.
9. Burattin, A., Maggi, F.M., Sperduti, A.: Conformance checking based on multi-
perspective declarative process models. Expert Systems with Applications 65
10. Carmona, J., van Dongen, B., Solti, A., Weidlich, M.: Conformance Checking:
Relating Processes and Models. Springer (2018)
11. Cenka, B.A.N., Santoso, H.B., Junus, K.: Analysing student behaviour in a learning
management system using a process mining approach. Knowledge Management &
E-Learning (2022)
12. Daniel, B.: Big Data and Analytics in Higher Education: Opportunities and Chal-
lenges: The Value of Big Data in Higher Education. Educational Technology 46(5)
13. Etinger, D.: Discovering and Mapping LMS Course Usage Patterns to Learning
Outcomes. In: IHSI. Springer (2020)
14. Heileman, G.L., Hickman, M., Slim, A., Abdallah, C.T.: Characterizing the Com-
plexity of Curricular Patterns in Engineering Programs. In: ASEE Annual Conf &
Exposition (2017)
15. Hilliger, I., Aguirre, C., Miranda, C., Celis, S., Pérez-Sanagustín, M.: Lessons
Learned from Designing a Curriculum Analytics Tool for Improving Student Learn-
ing and Program Quality. Computing in Higher Education (2022)
16. de Leoni, M., Maggi, F.M., van der Aalst, W.M.P.: Aligning Event Logs and Declar-
ative Process Models for Conformance Checking. In: BPM. Springer (2012)
17. Mannhardt, F., de Leoni, M., Reijers, H.A., van der Aalst, W.M.P.: Balanced
multi-perspective checking of process conformance. Computing 98(4) (2016)
18. Mathrani, A., Umer, R., Susnjak, T., Suriadi, S.: Data Quality Challenges in Ed-
ucational Process Mining: Building Process-Oriented Event Logs from Process-
Unaware Online Learning Systems. Business Information Systems 39(4) (2022)
19. Pechenizkiy, M., Trcka, N., de Bra, P., Toledo, P.A.: CurriM: curriculum mining.
In: EDM (2012)
20. Pechenizkiy, M., Trcka, N., Vasilyeva, E., van der Aalst, W., de Bra, P.: Process
Mining Online Assessment Data. International Working Group on Educational
Data Mining (2009)
21. Priyambada, S.A., Mahendrawathi, E.R., Yahya, B.N.: Curriculum Assessment
of Higher Educational Institution Using Aggregate Profile Clustering. Procedia
Computer Science 124 (2017)
22. Reiter, R.: A logic for default reasoning. Artificial Intelligence 13(1-2) (1980)
23. Salazar-Fernandez, J.P., Munoz-Gama, J., Maldonado-Mahauad, J., Bustamante,
D., Sepúlveda, M.: Backpack Process Model (BPPM): A Process Mining Approach
for Curricular Analytics. Applied Sciences 11(9) (2021)
24. Salazar-Fernandez, J.P., Sepúlveda, M., Munoz-Gama, J., Nussbaum, M.: Cur-
ricular Analytics to Characterize Educational Trajectories in High-Failure Rate
Courses That Lead to Late Dropout. Applied Sciences 11(4) (2021)
25. Schulte, J., Fernandez de Mendonca, P., Martinez-Maldonado, R., Buckingham
Shum, S.: Large Scale Predictive Process Mining and Analytics of University De-
gree Course Data. In: LAK ’17: Proceedings of the Seventh International Learning
Analytics & Knowledge Conference. Association for Computing Machinery (2017)
26. Srisamutr, A., Raruaysong, T., Mettanant, V.: A Course Planning Application for
Undergraduate Students Using Genetic Algorithm. In: ICT-ISPC (2018)
27. Sypsas, A., Kalles, D.: Reviewing Process Mining Applications and Techniques in
Education. Artificial Intelligence & Applications 13(1) (2022)
28. van der Aalst, W.M.P.: Process Mining: Data science in action. Springer (2018)
Process Mining and Rule-based AI for Study Planning and Monitoring 13
29. Wang, R., Zaïane, O.R.: Discovering Process in Curriculum Data to Provide Rec-
ommendation. In: EDM (2015)
30. Wong, C.: Sequence Based Course Recommender for Personalized Curriculum
Planning. In: AIED. Springer (2018)
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Online learning implementation has been growing year by year across countries, including Indonesia. Many higher education institutions use a Learning Management System (LMS) to facilitate online learning. Unfortunately, many issues arise during online learning implementation, such as a lack of student behaviour monitoring. This study adopts an educational process mining technique to conduct weekly assessments of student behaviour during one semester. The study was undertaken in the following steps: problem identification, literature review, design of study context, log data collection from LMS, log data filtering, event data grouping, conversion of LMS logs to event logs, clustering, and process model discovery. The following findings were revealed in this research: the most frequently accessed features were course material, assignments, and forums; students accessed the LMS most frequently on lecture days; the number of student activities decreased in line with fewer instructions from lecturers; students who attained the best grades most frequently accessed the LMS, and vice versa; and high-achieving students had a more complex process model than other students. Therefore, this research suggests that systematic teaching strategies have a broader impact on student engagement and performance.
Full-text available
Process Mining (PM) emerged from business process management but has recently been applied to educational data and has been found to facilitate the understanding of the educational process. Educational Process Mining (EPM) bridges the gap between process analysis and data analysis, based on the techniques of model discovery, conformance checking and extension of existing process models. We present a systematic review of the recent and current status of research in the EPM domain, focusing on application domains, techniques, tools and models, to highlight the use of EPM in comprehending and improving educational processes.
Full-text available
Late dropout is one of the most pressing challenges currently facing higher education, and the process that each student follows to arrive at that decision usually involves several academic periods. This work presents a curricular analytics approach at the program level, to analyze how educational trajectories of undergraduate students in high-failure rate courses help to describe the process that leads to late dropout. Educational trajectories (n = 10,969) of high-failure rate courses are created using Process Mining techniques, and the results are discussed based on established theoretical frameworks. Late dropout was more frequent among students who took a stopout while having high-failure rate courses they must retake. Furthermore, students who ended in late dropout with high-failure rate courses they must retake had educational trajectories that were on average shorter and less satisfactory. On the other hand, the educational trajectories of students who ended in late dropout without high-failure rate courses they must retake were more similar to those of students who graduated late. Moreover, some differences found among ISCED fields are also described. The proposed approach can be replicated in any other university to understand the educational trajectories of late dropout students from a longitudinal perspective, generating new knowledge about the dynamic behavior of the students. This knowledge can trigger improvements to the curriculum and in the follow-up mechanisms used to increase student retention.
The need to make default assumptions is frequently encountered in reasoning about incompletely specified worlds. Inferences sanctioned by default are best viewed as beliefs which may well be modified or rejected by subsequent observations. It is this property which leads to the non-monotonicity of any logic of defaults. In this paper we propose a logic for default reasoning. We then specialize our treatment to a very large class of commonly occuring defaults. For this class we develop a complete proof theory and show how to interface it with a top down resolution theorem prover. Finally, we provide criteria under which the revision of derived beliefs must be effected.
The widespread use of Learning Management Systems in higher education, a growing adoption of e-learning, distance, hybrid and blended learning puts much pressure onto both students, to achieve learning goals and lecturers, to design high quality online courses. Lecturers typically evaluate how much the students have achieved at the end of the course. This exploratory study attempts to uncover the relationship between usage behavior and students’ grades, i.e. what are the online course usage patterns performed by higher graded students in contrast to lower graded ones. The core data is of the analysis are the event logs extracted from the online course Modeling and simulation at the Faculty of informatics in Pula and mapped to the students’ grades accumulated from the final exams, assignments, projects and class tasks. Process mining techniques were used for process discovery and process model analysis. A set of procedures were developed (within the R programming environment) to analyze the discovered process models. The findings indicate that a better understanding of online course usage patterns and its relationship with learning outcomes can be used to develop intelligent systems (recommender systems, intelligent agents, intelligent personal assistants etc.) that can improve students’ learning process.
In this study, we develop and test four measures for conceptualizing the potential impact of co-enrollment in different courses on students’ changing risk for academic difficulty and recovery from academic difficulty in a focal course. We offer four predictors, two related to instructional complexity and two related to structural complexity (the organization of the curriculum) that highlight different trends in student experience of the focal course. Course difficulty, discipline of major, time in semester, and simultaneous difficulty across courses were all significantly related to entering a moderate and high-risk classification in the early warning system (EWS). Course difficulty, discipline of major, and time in semester were related to exiting academic difficulty classifications. We observe a snowball effect, whereby students who are experiencing difficulty in the focal course are at increased risk of experiencing difficulty in their other courses. Our findings suggest that different metrics may be needed to identify risk for academic difficulty and recovery from academic difficulty. Our results demonstrate what a more holistic assessment of academic functioning might look like in early warning systems and course recommender systems, and suggest that academic planners consider the relationship between course co-enrollment and student academic success.
Background: Process mining with educational data has made use of various algorithms for model discovery, principally Alpha Miner, Heuristic Miner, and Evolutionary Tree Miner. In this study we propose the implementation of a new algorithm for educational data called Inductive Miner. Method: We used data from the interactions of 101 university students in a course given over one semester on the Moodle 2.0 platform. Data was extracted from the platform's event logs; following preprocessing, the mining was carried out on 21,629 events to discover what models the various algorithms produced and to compare their fitness, precision, simplicity and generalization. Results: The Inductive Miner algorithm produced the best results in the tests on this dataset, especially for fitness, which is the most important criterion in terms of model discovery. In addition, when we weighted the various metrics according to their importance, Inductive Miner continued to produce the best results. Conclusions: Inductive Miner is a new algorithm which, in addition to producing better results than other algorithms using our dataset, also provides valid models which can be interpreted in educational terms.
Students in higher education need to select appropriate courses to meet graduation requirements for their degree. Selection approaches range from manual guides, on-line systems to personalized assistance from academic advisers. An automated course recommender is one approach to scale advice for large cohorts. However, existing recommenders need to be adapted to include sequence, concurrency, constraints and concept drift. In this paper, we propose the use of recent deep learning techniques such as Long Short-Term Memory (LSTM) Recurrent Neural Networks to resolve these issues in this domain.