OULAD MOOC Student Performance

Prediction using Machine and Deep Learning

Techniques

Wala Torkhani1∗ and Kalthoum Rezgui1,2

1
University of Manouba, ISAMM, University Campus of Manouba, Manouba,
Tunisia
2
University of Tunis, ISG of Tunis, SMART Lab, Tunis Tunisia
∗
[email protected]

Abstract. In online learning, the accurate prediction of student per-

formance is essential for timely interventions and personalized learning
experiences. This work leverages the Open University Learning Analyt-
ics Dataset (OULAD) to evaluate the effectiveness of various machine
learning (ML) and deep learning (DL) techniques in predicting student
performance. We implemented a range of models, including traditional
ML algorithms like Logistic Regression (LR), Decision Trees (DT), Ran-
dom Forests (RF), and Support Vector Machines (SVMs), as well as DL
models such as Convolutional Neural Networks (CNNs), Recurrent Neu-
ral Networks (RNN), and Long Short-Term Memory (LSTM) networks.
The performance of each model was assessed using various metrics such
as accuracy, precision, recall, and F1-score. The experimental results re-
vealed that, among DL models, LSTM prevailed in terms of accuracy
and precision, which are 83.41% and 82.20%, respectively. Additionally,
the RF and optimized DT models performed well and provided a strong
balance between accuracy and recall, making them a solid choice when
computational efficiency is a concern.

Keywords: Performance prediction, Learning Analytics, Machine Learn-

ing, Deep Learning, OULAD dataset

1 Introduction

During last years, predicting student performance has become increasingly im-
portant in the fields of educational data mining (EDM) [1] and learning analytics
(LA) [2]. Indeed, with the rise of online and blended learning environments, insti-
tutions are generating vast amounts of data on student interactions, behaviors,
and academic progress. Collecting this data to predict student success can en-
able educators and administrators to provide timely interventions, personalize
learning experiences, and improve overall educational outcomes.
In this context, the Open University Learning Analytics Dataset (OULAD) [3]
provides a rich source of data for exploring predictive models. It includes detailed

© The Author(s) 2025

S. Krichen et al. (eds.), Proceedings of International Conference on Decision Aid and Artificial Intelligence
(ICODAI 2024), Atlantis Highlights in Intelligent Systems 12,
https://doi.org/10.2991/978-94-6463-654-3_18
 OULAD MOOC Student Performance Prediction 229

records of student demographics, assessment results, and interactions with vir-

tual learning environments (VLE), making it an ideal dataset for evaluating the
efficacy of various ML and DL techniques in predicting student performance.
To predict student performance in online learning, ML techniques, like Logistic
Regression (LR), Decision Trees (DT), Random Forest (RF), and Support Vec-
tor Machines (SVMs), have been widely employed due to their ability to model
complex relationships within the data [10] [5].
While ML methods have proven effective, recent studies have also explored DL
approaches [12] [11]. In fact, the advent of deep learning has introduced more
sophisticated models, such as Convolutional Networks (CNNs) and Long Short-
Term Memory (LSTM) networks, which are capable of capturing intricate tem-
poral patterns and dependencies that may be missed by traditional ML algo-
rithms.
This paper aims to evaluate the performance of both traditional ML algorithms
and advanced DL models in predicting student performance using the OULAD
dataset. By comparing these approaches, we seek to identify the strengths and
limitations of each model and determine which one is most effective for this task.
The remainder of this paper is organized as follows: Section 2 provides a review
of related work in the field of student performance prediction. Section 3 presents
the OULAD dataset. Section 4 describes the methodology we proposed for pre-
dicting learners’ performance and outlines the ML and DL models implemented
in this study. Section 5 presents the experimental results and discussion. Finally,
Section 6 concludes the paper and suggests directions for future research.

2 Related Work

The prediction of student performance on OULAD dataset using ML and DL

techniques has recently attracted considerable attention in educational research,
with researchers employing a wide range of methods to enhance predictive ac-
curacy and interpretability. This section reviews key contributions to the field,
focusing on the application of traditional ML techniques and advanced DL mod-
els.

2.1 ML algorithms in student performance prediction

In recent years, a number of studies have employed ML techniques to analyze

the OULAD dataset, leveraging its rich information on student behavior and
performance.
In their work, [13] conducted a comprehensive study using various ML algo-
rithms, including LR, DT, and RF, to predict student performance and identify
at-risk students. Their findings highlighted the effectiveness of feature selection
in enhancing the predictive power of these models and provided insights into the
importance of different types of data for accurate predictions.
In [4], the authors focused on analyzing learner performance in digital learning
environments using ML methods to predict learners outcomes. The objective is
230 W. Torkhani and K. Rezgui

to develop indicators that alert students and teachers of potential challenges and
allowing a proactive support. The proposed methodology encompasses a trace
analysis approach, which involves manually and automatically selecting relevant
attributes followed by rule extraction to explain learner outcomes. While the DT
classifier is mentioned as one of the effective algorithms, the study also highlights
that Gaussian NB, KNN and LinearSVC classifiers are also among the top per-
formers based on precision, mean and standard deviation performance metrics.
An additional study conducted by [14] showed that ML algorithms achieved high
accuracy rates in predicting student performance, reaching 89.7% for KNN and
97.40% for SVM.
In [16], the authors reviewed several approaches for predicting student outcomes
in online courses using ML. They examined various ML techniques, including
DT, SVM, and NN, and evaluated their effectiveness in forecasting student out-
comes.

2.2 DL models in student performance prediction

With the advent of deep learning, new opportunities have been introduced for
modeling the intricate patterns present in educational data.
Artificial Neural Networks (ANNs), with their ability to learn hierarchical rep-
resentations, have been applied to various predictive tasks in education. For
instance, [6] employed ANNs to predict student performance, demonstrating
significant improvements in accuracy over traditional ML models, particularly
when dealing with large and complex datasets.
In [15], the authors used a transformer encoder model designed for predicting
student performance on the OULAD dataset. By analyzing students’ sequential
log activities, the model captures temporal dependencies and patterns in the
data, offering improved predictive accuracy, achieving approximately 83.17%,
compared to traditional methods.
Similarly, [11] proposed the DOPP model which leverages neural networks
(NN) to analyze various features, such as clickstream data and demographic
information, to predict student outcomes in online courses. This deep model
outperforms traditional machine learning methods (i.e., LR, SVM), demonstrat-
ing its ability to capture complex relationships in the data.
Recently, the Long Short-Term Memory (LSTM) networks, a type of recur-
rent neural network designed to capture temporal dependencies, have shown
promise in educational contexts where sequential data is critical.
In [7], the authors utilized LSTMs to model the progression of student knowl-
edge over time, achieving state-of-the-art results in predicting future perfor-
mance based on past interactions. This ability to model temporal sequences
makes LSTMs particularly suitable for analyzing clickstream data and other
time-dependent educational metrics.
Recent research has increasingly focused on hybrid approaches that combine
the strengths of both traditional ML and DL methods. These approaches aim
to leverage the interpretability of ML models with the powerful feature extrac-
tion capabilities of DL models. For instance, [9] developed a hybrid approach
 OULAD MOOC Student Performance Prediction 231

that integrates LR with LSTM networks to predict student dropout in MOOCs,

successfully balancing predictive accuracy with model interpretability.

3 OULAD Dataset
This section introduces the Open University Learning Analytics Dataset (OULAD)
(available for download here: https://analyse.kmi.open.ac.uk/open dataset) which
is one of the most popular open datasets in educational data mining and learning
analytics research [3].
The OULAD dataset encompasses data from seven different courses modules
(AAA GGG) and 22 courses offered by The Open University (OU) in the United
Kingdom. The data is collected from approximately 32,593 students who were
enrolled in these courses between the years 2013 and 2014. More precisely, this
dataset is structured across seven data tables (See Figure 1), each representing
different aspects of student interaction (clickstream data), demographic informa-
tion, and assessment outcomes. These tables are available in a comma separated

studentInfo studentAssessments assessments

studentRegistration
studentVle

courses

vle

Fig. 1. Detailed OULAD dataset structure.

value (CSV) format, the content of each .csv file is described below. Besides, the
corresponding details of these files have been provided in Table 1.
– Courses.csv: Contains metadata about each course, such as the module
code (code module) identifying the course, the code name of the presentation
(code presentation).
– Assessments.csv: Provides details on module presentations (code module,
code presentation) and various assessments within each course, including
the type of assessment (assessment type) (e.g., assignments, exams), and
the submission dates (date).
– StudentAssessment.csv: Contains the results of students’ assessments.
The date submitted attribute states the date of student submission, mea-
sured as the number of days since the start of the module presentation. The
232 W. Torkhani and K. Rezgui

is banked attribute is a status flag indicating that the assessment result has
been transferred from a previous presentation. The score column is the stu-
dent’s score in this assessment which ranges from 0 to 100. A score lower
than 40 is interpreted as Fail.
– studentInfo.csv: Includes demographic information of students, such as
age, gender, region of residence, highest education level, disability status,
socio-economic background, total number of credits for the module the stu-
dent is currently studying as well as the student’s final result (i.e., withdrawn,
fail, pass or distinction).
– StudentRegistration.csv: Contains the date when the students registered
for (and eventually unregistered from) a module presentation (date registration,
date unregistration). Students who completed the course have the field date unregistration
empty whereas students who unregistered have Withdrawal as the value of
the f inal result column in the studentInfo.csv file.
– StudentVle.csv: Captures students’ interactions with the VLE resources,
such as the number of clicks on different course materials, forum participa-
tion, and accessing to learning resources. The date attribute expresses the
date of student’s interaction with the material measured as the number of
days since the start of the module presentation. The sum click attribute
measures the number of times a student interacts with the material in that
day.
– Vle.csv: Includes information about the resources available in the VLE.

Table 1. CSVFiles and corresponding details in OULAD dataset.

File ID Attributes
courses.csv code module, code presentation, length
studentInfo.csv code module, code presentation,
id student, gender, region, high-
est education, imd band, age band,
num of prev attempts, studied credits,
disability, final result
studentRegistration.csv code module, code presentation,
id student, date registration,
date unregistration
studentAssessment.csv id student, id assessment, date submitted,
is banked, score
assessments.csv code module, code presentation,
id assessment, assessment type, date,
weight
vle.csv id site, code module, code presentation,
activity type, week from, week to
studentVle.csv id student, id site, code module,
code presentation, date, sum click
 OULAD MOOC Student Performance Prediction 233

4 Methodology
This section focuses on describing the methodology (See Figure 2) that we pro-
pose for predicting learners’ performance on the OULAD dataset using ML and
DL models.

Data collection
& OULAD
analysis Dataset

Data cleansing
Pre-processing and
preprocessing

Feature selection

Training Test
set set

K fold
cross ML & DL
validation training

Performance
Prediction

Fig. 2. Architecture of the proposed performance prediction approach.

4.1 Pre-processing
This stage encompasses the data pre-processing and transformation performed
prior to building the proposed predictive models. Initially, we filtered the original
dataset to focus on the AAA course module and the associated educational data
of 748 students who participated in this course. Then, the studentInfo table,
which contains demographic information, was merged with the studentRegistra-
tion table to track each student’s enrollment status, and with the studentAssess-
ment table to incorporate assessment scores [13]. Next, the studentAssessment
234 W. Torkhani and K. Rezgui

table was used to compute features related to students’ performance on indi-

vidual assignments and exams. These include total assessment scores, average
scores, the number of assessments submitted, and the the number of failed as-
sessments (assignment f ailed) for each student.
As the studentVle table contains detailed clickstream data recording every in-
teraction a student has with the VLE, this table is aggregated to create features
such as the number of unique sites visited, the total number of clicks, the number
of days active, and the average number of clicks per day.
The processed dataset is further cleaned by removing invalid entries while miss-
ing values were handled. For instance, numeric attributes, like as score, assign-
ment failed, id site, sum click, with missing values were filled with appropriate
replacements. In particular, missing score values were set to 0, indicating no
score was recorded, while missing assignment f ailed values were filled with the
maximum observed value, suggesting a high likelihood of failure.

4.2 Feature selection

As outlined in Section 3, the OULAD dataset consists of seven data tables with
40 unique columns (attributes). Among these attributes, we have selected and
added new attributes to learn and predict each learner’s final result based on the
data associated with these attributes. Indeed, attributes that are not significant
for the learning process have been removed, such as length, date unregistration,
id assessment, date submitted, is banked, week f rom, and week to.
Table 2 below provides a summary of the OULAD dataset after the feature
extraction step.

4.3 Proposed predictive models

This research work aims to use ML and DL methods to predict the performance
of students by analyzing their related features and interaction traces with the
learning environment. The main idea is to find the most appropriate prediction
model and compare the various performance metrics of each classifier used.

4.4 Machine Learning Models

In this research work, the following classification ML algorithms are used:

– Logistic Regression (LR) is a statistical model used for binary classification

tasks. It estimates the probability of a binary outcome (e.g., pass/fail) based
on one or more predictor variables.
– Decision Trees (DT) are a type of supervised learning model that splits the
data into subsets based on feature values to make predictions.
– Decision Tree Optimization (DT opt) is an optimized DT model that in-
volves tuning the hyperparameters of the C4.5 algorithm to improve its per-
formance.
 OULAD MOOC Student Performance Prediction 235

– Random Forest (RF) is an ensemble learning method that constructs mul-

tiple decision trees during training and outputs the mode of the classes (for
classification).
– Support Vector Machines (SVMs) is a supervised learning model used pri-
marily for classification tasks. It works by finding the optimal hyperplane
that maximizes the margin between different classes in the feature space.
– Naive Bayes (NB) is a probabilistic classifier based on Bayes’ theorem. It
assumes that the features are conditionally independent given the class label
(hence ”naive”).

4.5 Deep Learning Models

The following DL models were considered for predicting student performance:

– Multilayer perceptron (MLP) is a class of feedforward artificial neural net-

works consisting of multiple layers of neurons. It typically includes an input
layer, one or more hidden layers, and an output layer, where each layer is
fully connected to the next.
– Convolutional Neural Networks (CNNs) are primarily used for image pro-
cessing but can be adapted for structured data by treating data as a grid.
They apply convolutional layers to extract features from input data. CNNs
are effective at identifying hierarchical patterns and reducing dimensionality.
– Recurrent Neural Networks (RNNs) are designed to handle sequential data
by maintaining a memory of previous inputs. Their main advantage lies in
their ability to handle sequential and time-series data, capturing dependen-
cies over time.
– Long Short-Term Memory (LSTM) is a type of RNN specifically designed
to overcome the limitations of traditional RNNs by maintaining long-term
memory through a series of gates that control the flow of information. These
networks are used to predict student performance by analyzing sequences of
interactions over time, allowing the model to learn temporal dependencies.

5 Experiment and results

In this section, we present the experimental results of the proposed ML and DL

models. First, we outline the parameter settings used for each ML and DL model
applied in this work. Then, we give the experimental results for these models
in predicting students’ performance. The implementation and experiments of
all proposed models were implemented using the Python 3.0 language and its
Scikit-Learn and TenserFlow libraries.

5.1 Parameter settings

Following preprocessing, the dataset was split into two data sets with a ratio of
7:3, i.e., 70% of data was used as training set and the rest 30% of the data was
236 W. Torkhani and K. Rezgui

used as testing set.

The evaluation metrics we selected to assess the performance of the proposed
prediction models are accuracy, precision, recall, and F1-score.
In the model development stage, each model was configured with a set of hyper-
parameters that influence the learning process and the resulting performance.
These hyperparameters were either set to their default values or tuned based on
cross-validation to improve the models’ performance on the OULAD dataset.
The following table (see Table 3) presents the parameters used for each model,
along with their descriptions and values.

5.2 Prediction results

Figure 3 illustrates the comparative performance of proposed ML models across

the key performance metrics. It can be seen from this figure that the RF model
achieved the highest scores for most metrics, particularly excelling in recall and
accuracy with values of 85.69% and 82.73%.
It can, also, be revealed that the DT opt and LR models performed well, their
accuracy are between 78% and 79%. However, DT opt provides a better recall
than LR, making it a solid choice to get accurate predictions. The NB model
is the less effective in terms of all performance metrics, but still offers value for
rapid and lightweight predictions.

Fig. 3. Comparative performance analysis of ML models.

The experimental results in Figure 4 illustrates the performance of various deep

learning models in predicting student performance.
The LSTM model exhibits the highest performance across all metrics, with an
accuracy of 83.41%, a precision of 82.20%, and a recall of 81.88%. This indicates
that the LSTM model is the most effective in modelling time-series data and
sequential nature of the OULAD dataset.
 OULAD MOOC Student Performance Prediction 237

Fig. 4. Comparative performance analysis of DL models.

6 Conclusion

In this study, we explored the effectiveness of various machine learning and deep
learning models in predicting student performance using the OULAD dataset.
Our goal was to identify the most suitable model for accurately predicting stu-
dent outcomes, which is crucial for enabling timely interventions and enhancing
educational support systems.
The results of our experiments indicate that DL models, particularly LSTM,
outperforms other models in accuracy, precision, and recall metrics.
Among traditional ML models, RF and DT opt demonstrated strong perfor-
mance, making them viable alternatives, especially when computational effi-
ciency or model interpretability is a priority.
The significance of these findings advances that educational institutions can
leverage LSTM models to better predict student performance in online learn-
ing and MOOCs. Furthermore, while DL models require more computational
resources, their superior performance in handling complex and sequential data
justifies their use in scenarios where prediction accuracy is crucial.
Future research could focus on exploring hybrid approaches that combine the
strengths of both ML and DL techniques, and expanding the analysis to include
additional datasets and educational environments. Moreover, integrating inter-
pretability techniques will be essential to ensure that the predictions made by
DL models can be understood and acted upon by educators.
238 W. Torkhani and K. Rezgui

Table 2. Description of the OULAD dataset after feature selection.

Attributes Description
code module AAA Identification code of AAA module course on which the student
is registered.
code module DDD Identification code of DDD module course on which the student
is registered.
code presentation Code name of the presentation
age band 0-35, age band 35-55 Different age ranges
assignment failed Number of assignments failed by a student
sum click x Sum of the students clicks on online learning materials before
the first third of the course module
sum click y Sum of the student’s clicks on online learning materials during
the remaining period
imd band 0-10%, Different bands of the Index of multiple deprivation
imd band 10-20%,
imd band 20-30%,
imd band 30-40%,
imd band 40-50%,
imd band 50-60%,
imd band 60-70%,
imd band 70-80%,
imd band 80-90%
gender F Indicator or binary flag for female gender
gender M Indicator or binary flag for male gender
id student A unique identification number for the student
studied credits Total number of credits for the modules
date registration Number of days measured relative to the start of the module
presentation
num of prev attempts Number of times the student has attempted the course module
score Student’s score in an assessment
date diff Difference in days between the planned date for an assessment
and the date submitted
activity type Type of activity in the VLE converted into a numerical index
total days Total number of noted activity days
actionArray train and Students interactions over time with the VLE encoded using
actionArray test three-dimensional arrays representing the number of clicks per
student, per day, and per activity type
region East Anglian, Variables for different regions
region East Midlands,
region Ireland, region South,
region London,
region North, region Scotland,
region North Western,
region South East,
region South West,
region Wales,
region West Midlands
 OULAD MOOC Student Performance Prediction 239

Table 3. Prediction models parameter settings.

Model Parameter Description Value

LR C Inverse of regularization strength 0.1
max iter Maximum number of iterations taken for the solvers to 100
converge
penalty A regularization term added to the loss function to pre- l1
vent overfitting
solver Algorithm to use in the optimization problem saga
SVM C Regularization parameter strength 1.0
kernel Kernel type to be used in the algorithm rbf
DT max depth Maximum depth of the tree 7
min samples split Minimum number of samples required to split an internal 5
node
criterion Function to measure the quality of a split gini
RF n estimators Number of trees in the forest 100
max features Number of features to consider when looking for the best auto
split
bootstrap Whether bootstrap samples are used when building trees true
NB var smoothing Portion of the largest variance of all features added to 1e-9
variances for numerical stability
MLP activation Function applied to the output of each neuron in a layer sigmoid
Number of neurons in each hidden layer
hidden units Number of time steps that the model processes at once 128
n steps Number of input features that the model processes at 10
n features each time step or input instance 20
Function used to measure the difference between the
loss model’s predictions and the actual target values binary
Algorithm used to update the model’s weights during crossen-
optimizer training in order to minimize the loss function tropy
Adam
CNN activation Function applied to the output of each neuron in a layer sigmoid
Number of neurons in each hidden layer
hidden units Number of time steps that the model processes at once 10
n steps Number of filters in the convolutional layers 20
filters Size of the convolutional kernel 64
kernel size Size of the max-pooling window (3,3)
pool size (2,2)
RNN units Number of LSTM units (neurons) in the layer 64
activation Function applied to the output of each neuron in a layer ReLU

LSTM units Number of LSTM units (neurons) in the layer 300

n steps Number of time steps or sequential inputs that the model 10
processes
n features Number of input features that the model processes at 20
each time step
mask value A value that is ignored or masked during training or eval- 0
uation
return sequences Specifies whether the output of each time step should be False
returned or just the output of the final time step
dropout Fraction of the input units to drop for regularization 0.1
recurrent dropout Fraction of the recurrent units to drop for regularization 0.1
Function applied to the output of each neuron in a layer
activation sigmoid
 OULAD MOOC Student Performance Prediction 240

References

1. Peña-Ayala, A. Educational data mining: A survey and a data mining-based anal-

ysis of recent works. Expert Systems With Applications. 41, 1432-1462 (2014)
2. Leitner, P., Khalil, M. & Ebner, M. Learning analytics in higher education—a
literature review. Learning Analytics: Fundaments, Applications, And Trends: A
View Of The Current State Of The Art To Enhance E-learning. pp. 1-23 (2017)
3. Kuzilek, J., Hlosta, M. & Zdrahal, Z. Open university learning analytics dataset.
Scientific Data. 4, 1-8 (2017)
4. Sehaba, K. Learner Performance Prediction Indicators based on Machine Learning.
CSEDU (1). pp. 47-57 (2020)
5. Albreiki, B., Zaki, N. & Alashwal, H. A systematic literature review of stu-
dent’performance prediction using machine learning techniques. Education Sci-
ences. 11, 552 (2021)
6. Kotsiantis, S., Pierrakeas, C. & Pintelas, P. Predictiong students’ performance in
distance learning using machine learning techniques. Applied Artificial Intelligence.
18, 411-426 (2004)
7. Piech, C., Bassen, J., Huang, J., Ganguli, S., Sahami, M., Guibas, L. & Sohl-
Dickstein, J. Deep knowledge tracing. Advances In Neural Information Processing
Systems. 28 (2015)
8. Crossley, S., Paquette, L., Dascalu, M., McNamara, D. & Baker, R. Combining
click-stream data with NLP tools to better understand MOOC completion. Pro-
ceedings Of The Sixth International Conference On Learning Analytics & Knowl-
edge. pp. 6-14 (2016)
9. Whitehill, J., Williams, J., Lopez, G., Coleman, C. & Reich, J. Beyond prediction:
First steps toward automatic intervention in MOOC student stopout. Available At
SSRN 2611750. (2015)
10. Chen, Y. & Zhai, L. A comparative study on student performance prediction us-
ing machine learning. Education And Information Technologies. 28, 12039-12057
(2023)
11. Karimi, H., Huang, J. & Derr, T. A deep model for predicting online course perfor-
mance. Association For The Advancement Of Artificial Intelligence. pp. 1-6 (2020).
12. Alnasyan, B., Basheri, M. & Alassafi, M. The Power of Deep Learning Techniques
for Predicting Student Performance in Virtual Learning Environments: A Sys-
tematic Literature Review. Computers And Education: Artificial Intelligence. pp.
100231 (2024)
13. Gašević, D., Dawson, S., Rogers, T. & Gasevic, D. Learning analytics should not
promote one size fits all: The effects of instructional conditions in predicting aca-
demic success. The Internet And Higher Education. 28 pp. 68-84 (2016)
14. Qiu, F., Zhang, G., Sheng, X., Jiang, L., Zhu, L., Xiang, Q., Jiang, B. & Chen,
P. Predicting students’ performance in e-learning using learning process and be-
haviour data. Scientific Reports. 12, 453 (2022)
15. Kusumawardani, S. & Alfarozi, S. Transformer encoder model for sequential pre-
diction of student performance based on their log activities. IEEE Access. 11 pp.
18960-18971 (2023) .
16. Alhothali, A., Albsisi, M., Assalahi, H. & Aldosemani, T. Predicting student out-
comes in online courses using machine learning techniques: A review. Sustainability.
14, 6199 (2022).
 OULAD MOOC Student Performance Prediction 241

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-
NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/),
which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any
medium or format, as long as you give appropriate credit to the original author(s) and the
source, provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter's
Creative Commons license, unless indicated otherwise in a credit line to the material. If material
is not included in the chapter's Creative Commons license and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain
permission directly from the copyright holder.