www.ijcrsee.com
175
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
Original scientific paper
Received: December 10, 2024.
Revised: March 21, 2025.
Accepted: April 11, 2025.
UDC:
004.42:004.738.5
004.855.6
10.23947/2334-8496-2025-13-1-175-190
© 2025 by the authors. This article is an open access article distributed under the terms and conditions of the
Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
*
Corresponding author: stamenkovic.srecko@gmail.com
Abstract: In this paper, we present a web application designed for learning and visualizing Support Vector Machine
(SVM) algorithms, which are key components in the fields of machine learning and data processing. The application was
developed as an interactive tool that allows students and researchers to experiment with SVM models, providing insight into
their structure and functionality. By using modern web technologies, the application offers a user environment that is accessible,
intuitive, and adaptable for learning and research. In addition to implementing a web tool for learning the SVM algorithm, this
study proposes a method for its application in teaching and analyzes the impact of applying the new interactive method on final
learning outcomes. To assess the effectiveness of this tool, an experiment was conducted consisting of three phases: pre-testing,
training, and post-testing. To evaluate students’ experiences with the applied alternative learning method using the auxiliary
tool and their perception of the software system’s effectiveness, the standardized System Usability Scale (SUS) was used.
Keywords: Support Vector Machine, educational technology, algorithm visualization, simulation systems, educational tools.
Nenad Jovanović
1
, Stefan Jovanović
2
, Srećko Stamenković
3
, Dražen Marinković
4
,
Negovan Stamenković
5
1
Faculty of Technical Sciences, University of Pristina, Kosovska Mitrovica, Serbia, e-mail: nenad.jovanovic@pr.ac.rs
2
Faculty of Electronic Engineering, University of Nis, Serbia, e-mail: stefan.n.jovanovic@elfak.rs
3
Toplica Academy of Applied Studies, Department of Business Studies Blace, Serbia, e-mail: stamenkovic.srecko@gmail.com
4
University MB, Belgrade, Serbia, e-mail: drazenmarinkovic@gmail.com
5
Faculty of Sciences and Mathematics, University of Pristina, Kosovska Mitrovica, Serbia,
e-mail: negovan.stamenkovic@pr.ac.rs
A Web Application for Learning Support Vector Machine Algorithms in
Computer Engineering
Introduction
With the development of machine learning, intelligent software systems are increasingly being de-
veloped today, which, based on input data, use one or a group of algorithms to provide appropriate output
data as a result of execution. Machine learning algorithms are used in a wide range of different fields,
including medical diagnostics, control of robotic systems, management of industrial systems, telecommu-
nications, finance and stock trading, the computer game industry, the music industry, and many others.
However, in many cases, machine learning models are considered black boxes because the inner func-
tionality of the underlying algorithms is not entirely understandable to analysts (Mühlbacher et al., 2014),
and even for experts, tuning and parameterizing certain models can be challenging (Zeiler and Fergus,
2014). Due to the complexity of machine learning algorithms, there is a need for systems that enable their
application in everyday work as well as in the education process. Well-known software systems, environ-
ments, and libraries for research and concrete application of machine learning algorithms are Weka (Hall
et al., 2009), TensorFlow (Abadi et al., 2015), PyTorch (Paszke et al., 2019), Scikit-learn (Pedregosa et
al., 2011), Keras (Chollet et al., 2015), and Shogun (Sonnenburg et al., 2010). The focus of this paper is
on systems used as support in education to help students understand the dynamic behaviors of machine
learning algorithms. The analysis of existing educational tools for visualization and simulation of machine
learning was conducted in the second section.
www.ijcrsee.com
176
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
The goal of machine learning is to construct algorithms capable of learning to predict specific target
outputs. To achieve this, the learning algorithm is given specific training examples that demonstrate the
target relationship between input and output values. The algorithm should generate approximately correct
output, including examples that it did not encounter during training. For a good understanding of these
topics, undergraduate students need to master other areas of computer engineering, and it is desirable
to have some prior knowledge, which is often not acquired at previous levels of education (Çağlayan,
2019). Therefore, it is very important to properly motivate students to learn these topics. Some of the
pedagogical methods recommended for teaching machine learning are hands-on tasks, collaborative
learning, and problem-based learning (Hazzan and Mike, 2023). Hands-on tasks involve engaging stu-
dents in real hands-on projects that include data collection, model training, and performance evaluation.
This encourages practical application and solving the set tasks. Through the analysis of real problems,
Problem-Based Learning helps students develop critical thinking and practical problem-solving skills. Col-
laborative learning allows students to solve problems together, share ideas and learn from each other.
Computer engineering continuously strives for innovations in pedagogical methods and tools, especially
in the field of machine learning. Auxiliary software tools should enable students and future engineers to
visualize and simulate different algorithms for educational purposes and to gain insight into the possible
effects of certain algorithms, thus more easily selecting the appropriate algorithm for specific needs and
optimally determining the parameters with which to use the appropriate algorithms through simulation.
Rutten et al. (2012) considered a large number of experimental studies on the effects of software
tools in teaching published over a decade. Many studies reviewed in that work compare working condi-
tions with and without auxiliary tools, showing positive results in favor of using software tools to improve
traditional learning methods. Today, it is common to use different learning support tools in computer
engineering courses during laboratory exercises. In courses more oriented towards studying hardware
and abstract systems, such tools are often software simulators (Djordjevic et al., 2005; Jovanovic et al.,
2012; Stamenkovic and Jovanovic, 2024) whose task is to simulate the system being studied. In software
courses, tools for visual representation of algorithms are often used (Thakur et al., 2011; Stamenkovic et
al., 2023). The field of designing tools for the visual representation of algorithms has a long tradition, and
many such software systems have been realized so far. Interpreting machine learning models is currently
a popular topic in the information visualization community, and the results demonstrate that insights from
machine learning models can lead to better predictions and improved reliability of results (Chatzimparm-
pas et al., 2020). With the growing popularity of machine learning, there is also a greater need for systems
for the visual representation of algorithms in this field.
Support Vector Machine (SVM) is one of the fundamental algorithms in the field of machine learn-
ing, and understanding it is crucial for a deep comprehension of the theoretical and practical aspects
of this field. Support Vector Machines (SVM) are powerful machine learning algorithms used for clas-
sification, regression, and anomaly detection (Cortes and Vapnik, 1995; Bennett and Campbell, 2000;
Schölkopf and Smola, 2002). The idea of the method is to find a separating hyperplane in the vector space
where the data are represented so that all data from the same class are on the same side of the plane,
as shown in Figure 1. SVM finds the optimal separating hyperplane, which is the plane with the maximum
margin. The margin represents the width of the separation between classes that should be maximized.
SVM algorithms are characterized by their ability to effectively work in high-dimensional spaces and with
data that are not linearly separable due to the use of kernel functions. Since SVM effectively classifies
multidimensional data of various types, it is successfully applied in various fields, including biosciences
for analyzing biological processes, medicine for analyzing medical images, DNA analysis, and predicting
population structure. It is also widely used for classifying hate speech on social networks. Due to their
wide application and significance, SVM algorithms are an integral part of university course literature on
machine learning.
www.ijcrsee.com
177
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
Figure 1. An example showing the optimal hyperplane with the maximum margin that separates data into two
classes
The aim of this study is the development of a web application that serves as an educational tool for
learning and visualizing the principles and applications of SVM algorithms. The application is developed
using modern web technologies such as HTML-CSS, JavaScript, and Java EE technologies, enabling
interactive and visual learning. The use of these technologies contributes to the flexibility and accessibility
of the application, making it available to a wide range of users, including students and researchers.
The paper is systematized into seven sections. Following the introduction, the second section ana-
lyzes relevant literature to present existing solutions in this field. The third section describes the system
architecture with an emphasis on modular design, integration with databases, and how SVM processing is
performed within the application. A detailed description of the user interface, with examples of how users
interact with the application, including the visualization of SVM models, is presented in the fourth section.
The fifth section presents a case study that precisely describes the strategy for learning the SVM algo-
rithm using the developed auxiliary tool. The sixth section provides a detailed description of the conducted
experiment and the results of the tool evaluation, along with a discussion. Conclusions and suggestions
for future work are presented in the seventh chapter.
Literature Review
Learning support tools are part of the modern educational process. The main task of these systems
is to enable users to master the material more easily. Nowadays, there are a large number of diverse
learning support systems. Depending on the field of application and the level of knowledge required to
use the system, the technology applied for system implementation and the didactic methods that the
systems implement differ. For supporting the learning of algorithms in the fields of machine learning and
artificial intelligence, visualization and simulation tools for algorithm operation are successfully used. The
perception of reality is fundamentally visual, and in many cases, humans use symbolic processing, visual
diagrams, and other forms of imaginative processes to gain intuition about what, in its formal aspect,
has an abstract structure (DÍaz, Dormido, and Rivera, 2015). Today, it is widely accepted that visualizing
information about the internal workings of complex algorithms can contribute to better understanding and
offer solutions for better and more interpretable machine learning models (Sacha et al., 2017). Below,
some of the systems developed as tools to support learning algorithms in the fields of machine learning
and artificial intelligence will be presented.
Principal Component Analysis (PCA) is the most commonly used procedure in exploratory data
analysis and in machine learning for predictive models. However, due to its complexity, the mechanisms
and results of this operation are difficult for users to understand. To help students better understand and
use PCA, Jeong et al. (2009) developed a system called iPCA (interactive PCA), which visualizes the
results of principal component analysis using multiple different views and a rich set of user interactions.
The interactive iPCA system with an example is shown in Figure 2. As can be seen, the visualization is
realized at a high level through multiple different views, and there are options for configuring the param-
eters of the operation.
www.ijcrsee.com
178
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
Figure 2. Interactive iPCA System: (A) Projection View; (B) Characteristic Vectors Overview; (C)
Data View; (D) Correlation Overview; (E) Dimension Control; (F) Adjustment Options.
Mayfield and Rosé (2010) focus on the development of an interactive tool that aids in the analysis of
errors in the text retrieval process. The proposed tool is designed to enable users, especially students and
researchers, to more easily identify and understand errors that occur during text classification. The paper
emphasizes the importance of understanding errors in machine learning models and presents a user inter-
face (Figure 3) that allows for visual exploration and analysis of errors. The proposed tool is useful for edu-
cation because it provides deeper insight into the decision-making processes of machine learning models.
Figure 3. Interface of the error analysis tool in the text retrieval process
Kim et al. (2021) presented an interactive framework for demonstrating symbolic regression that
allows users not only to perform general configuration but also to control the system during training. The
demonstration system for symbolic regression enables user interaction with the DSR (deep symbolic re-
gression) algorithm. The interface provides real-time visualization and diagnostics to help the user monitor
and control the algorithm. The interactive platform includes visualization of top-performing mathematical
expressions and real-time algorithm diagnostics. This software system consists of a core algorithm (DSR)
running in the background, a framework for algorithm visualization, and an adaptive user interface for data
www.ijcrsee.com
179
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
loading and real-time configuration (Figure 4). The interactive platform is web-based, meaning that the
user interacts with the algorithm through a web browser while execution is performed on the server side.
Figure 4. User interface of the symbolic regression demonstration tool
NNeduca (Jovanovic et al., 2023) is a software environment designed for teaching and learning the
principles of artificial neural networks. The tool was created with the intention of improving both teaching
methods and learning outcomes. The proposed system can be used to design a neural network with any
number of layers and an arbitrary number of neurons in each of those layers. The development process of
the neural network is flexible and configurable, the technical features and appearance of the environment
are customizable, and the choice of transfer function for each layer is optional. All processes in neural net-
works are presented visually, and the results are illustrated using appropriate graphs. Neural networks can
be applied to both regression and classification tasks, making this tool applicable for various calculations in
many fields. The system focuses on students for whom studying neural networks is a challenging task, pro-
viding them with the opportunity to experiment with different types of neural networks, adjust parameters,
train networks on various datasets, and visually analyze the results through an intuitive user interface.
Figure 5. User interface of the NNeduca software environment
No tool developed at any university for assisting in the learning of SVM algorithms was found in
the literature search. This fact served as an additional motivation for the development of such a system.
www.ijcrsee.com
180
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
System architecture
Figure 6 shows the architecture of the software system. The basic components of the system are:
1. Frontend (HTML, CSS, JavaScript):
• HTML, CSS, JavaScript: The user interface allows the input of parameters for the SVM model.
• AJAX: Enables asynchronous data sending and receiving of results without refreshing the
page.
• Visualization System.
2. Backend:
• Java EE: Enterprise application for handling user requests and business logic.
• SvmServlet: Receives AJAX requests with parameters, forwards them to SvmProcessorBean,
and returns results as JSON.
• SvmProcessorBean: Processes data, trains the SVM model using libsvm, and returns results
to the servlet.
3. Libraries:
• libsvm: Library for implementing SVM algorithms for model training.
• D3.js: Visualizes results through interactive graphs.
Figure 6. System architecture
The user configures parameters for the SVM model via the web interface (frontend) by entering
values such as the type of SVM, kernel function, C, gamma, and other parameters. Once the parameters
are set, an asynchronous HTTP request is sent to the SvmServlet (backend) using AJAX, which provides
a smooth user experience without requiring page reloading. The SvmServlet receives these parameters
and forwards them to the SvmProcessorBean component. The SvmProcessorBean uses the libsvm li-
brary to train the SVM model with the given parameters. After processing, the SvmProcessorBean returns
the results to the SvmServlet, which then sends a JSON response with the results back to the frontend
via AJAX. The results are then displayed on the frontend using D3.js to create interactive graphs, which
allow the user to visualize the performance of the trained model.
This architecture allows users to interactively experiment with SVM models using the flexibility of
AJAX requests for quick processing and result visualization. All layers of the system are clearly defined
and interconnected, allowing for efficient and clear handling of user requests and result visualization.
Practical work with the system
The page at https://ml.pr.ac.rs/svm.html provides an interactive environment for working with Sup-
port Vector Machine (SVM) models (Figure 7). It allows users to experiment with different types of SVMs
(C-SVC, nu-SVC, one-class SVM, epsilon-SVR, nu-SVR) and kernel functions (Linear, Polynomial, Radial
Basis Function, Sigmoid).
www.ijcrsee.com
181
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
Figure 7. The main window of the application
Users can configure parameters for the SVM model via the web interface. The data are then loaded
into the DATA SET module. The data is in Excel format. It is necessary to specify which columns in the
Excel table represent the input data (in this case, 0,1), while the remaining columns represent the output
data. In addition, the percentage of data to be used for testing must be specified (in this case, 10%). After
that, by clicking the LOAD DATA button, the data will be loaded (Figure 8).
Figure 8. Data set module
The data and parameters are then sent asynchronously via an HTTP request to the backend (Svm-
Servlet) using AJAX. On the backend, the request is processed. SvmServlet receives the parameters and
forwards them to SvmProcessorBean.
SvmProcessorBean is an Enterprise JavaBean (EJB) component used to implement business logic
in the SVM visualization application. Its main functions include processing parameters, training the SVM
model, and generating results. SvmProcessorBean returns the results to SvmServlet.
SvmServlet is a Java Servlet component that plays a key role in the communication between the
frontend and the backend of the application. Its main functions include receiving parameters, forwarding
parameters, processing results, and sending the response back to the frontend. SvmServlet sends a
JSON response with the results back to the frontend via AJAX. The results are displayed on the frontend
using D3.js to create interactive graphs (Figure 9).
Different types of SVMs allow the algorithm to be applied to a wide range of tasks, from classifica-
tion to regression and anomaly detection, providing flexibility and efficiency in solving various machine
learning problems.
The user selects one of several types of SVMs. C-SVC (C-Support Vector Classification) is a clas-
sic SVM for classification. It uses regularization with the parameter C, which controls the balance between
maximizing the margin and minimizing classification errors. An alternative is nu-SVC (nu-Support Vector
Classification) with the parameter ν (nu), which controls the upper bound on the fraction of errors and sup-
port vectors, providing a more flexible approach to controlling model complexity. A one-class SVM is used
for anomaly detection. Training is performed only on positive examples, and the model identifies whether
new data belongs to the same class. Epsilon-SVR (Epsilon-Support Vector Regression) is a regression
SVM for predicting continuous values. The parameter epsilon controls the width of the region around the
www.ijcrsee.com
182
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
function within which errors are not penalized. Nu-SVR (nu-Support Vector Regression) is a regression SVM
with parameters ν (nu) and epsilon for better control over the number of support vectors and error tolerance.
Kernel functions are mathematical functions that transform data into a higher dimension so that the
SVM algorithm can more easily find a separating hyperplane for classification or regression. They enable
the SVM to model nonlinear boundaries between classes. In the application, the following kernel functions
can be used:
1. Linear: Uses a linear kernel, the simplest function for separating data. It is efficient for high-
dimensional data.
2. Polynomial: Uses a polynomial kernel that can model nonlinear relationships. Parameters in-
clude the degree of the polynomial and the coefficient.
3. Radial Basis Function (RBF): The most commonly used kernel for SVM. The parameter gamma
controls the width of the RBF and adjusts the flexibility of the model.
4. Sigmoid: Uses a sigmoid kernel, which is similar to the neuron activation function. It can model
nonlinear relationships and is used in support vector networks.
Within the SVM (Support Vector Machine) model, there are several parameters that can be ad-
justed to optimize performance. Understanding these parameters is crucial for effectively using SVM al-
gorithms in machine learning. Below is a detailed description of the most important adjustable parameters
for SVM models, explaining their function and impact on the model:
1. C (Regularization Parameter): Controls the trade-off between achieving a low training error and
a low testing error, which is achieved by maximizing the margin.
2. Gamma (Parameter for RBF, Polynomial, and Sigmoid Kernels): Controls the width of the
Gaussian curve in the RBF kernel. Lower gamma values create wider curves (smoother bounda-
ries), while higher values create narrower curves (more precise boundaries).
3. Cachesize: Determines the amount of memory allocated for caching the kernel matrix. Larger
values can speed up the computation of the kernel matrix but require more memory.
4. Epsilon (Parameter for Epsilon-SVR): Controls the width of the epsilon-tube within which er-
rors are not penalized. Larger values of epsilon make the model more tolerant of errors within the
epsilon-tube.
5. Nu (Parameter for nu-SVC and nu-SVR): Controls the upper bound on the fraction of errors and
support vectors. It allows direct control of model complexity and error tolerance.
6. Coef0 (Parameter for Polynomial and Sigmoid Kernels): A constant added in the polynomial
and sigmoid kernels. It affects the shape and complexity of the model when using nonlinear ker-
nel functions.
7. Degree (Degree of Polynomial for Polynomial Kernel): Determines the degree of the polyno-
mial in the polynomial kernel. Higher values make the model more complex and allow it to capture
nonlinear patterns in the data.
Users can adjust these parameters to optimize the performance of their SVM models for specific
problems. By experimenting with different values of these parameters, it is possible to achieve a balance
between model accuracy and its generalization (Figure 9).
Figure 9. SVM model results visualization
www.ijcrsee.com
183
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
The obtained model can be used for classification by clicking the USING MODEL button (Figure
10).
Figure 10. Test data
The data should be in Excel format, and the classification results should be obtained in graphical
form (Figure 11).
Figure 11. Classification results
This application significantly contributes to education through its interactive and practical approach.
By providing users with the ability to directly experiment with different SVM models and parameters, the
application enables a deeper understanding and efficient application of theoretical knowledge in practice.
With its rich visualizations and flexibility, the application is an invaluable tool for students, researchers,
and professionals in the field of machine learning.
Case study
The presented web-based tool is suitable for supporting the learning/teaching of SVM algorithms in
all undergraduate courses that cover machine learning algorithms. The alignment of this software system
with the literature used in theoretical teaching is of exceptional importance due to its usability and easier
integration within laboratory exercises. Given that the tool covers different types of SVMs as well as all
kernel functions, it will be suitable for many undergraduate curricula. For learning complex theoretical
constructs, the active learning method is recommended (Silva et al., 2019; Pradono et al., 2013), and this
method is also applied when learning SVM through the auxiliary tool. Collaboration among students also
has a significant impact on learning outcomes (Guo et al., 2020). The learning outcomes of applying the
realized educational tool include: (1) understanding how the SVM algorithm works; (2) identifying its ad-
vantages and disadvantages compared to other classification algorithms; (3) the ability to recognize prob-
lems for which the SVM algorithm is most suitable; (4) acquiring the skill to correctly select the appropriate
type of SVM, kernel function, and properly configure it for the most efficient solution to the given problem.
www.ijcrsee.com
184
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
For learning this algorithm, 4 hours of theoretical and 4 hours of practical instruction are recom-
mended. The teaching should be based on the traditional paradigm of theory-examples-exercises. Thus,
the strategy for learning the SVM algorithm by applying the developed auxiliary tool is as follows:
• Theoretical concepts are first presented for each topic during the theoretical lectures. (Auxiliary tools
for visualization and simulation are powerful teaching aids only when used in conjunction with tradi-
tional teaching (Taher and Khan, 2014).
• Laboratory exercises involve a more detailed explanation of the topics covered in lectures using tools
to visually represent the presented concepts. To prepare for the exercise, students should read the rel-
evant lecture material and textbooks, as well as the accompanying laboratory material. Each exercise
consists of two components:
• Demonstration of Examples: At the beginning of the laboratory exercise, the instructor demon-
strates examples using the educational tool. The instructor should utilize the system’s potential
to engage students, as otherwise, the exercises would merely represent a different approach
to theoretical presentation. In line with the principles of active learning, an interactive approach
to the demonstration involves engaging students so that they are not just passive observers.
The easiest way to include students during the demonstration is by frequently asking “what if”
questions and prompting them to predict the results.
• Laboratory Assignment: After the demonstration is completed, students receive appropriate
laboratory tasks that they complete independently.
By applying this strategy, students show a high level of motivation and actively participate in the en-
tire process and discussion. Another advantage of the presented web tool is that once students become
familiar with it, they can use it independently after class as an additional resource for solving homework
assignments and seminar papers.
Experiments and evaluation
This study employs a descriptive research design that relies on quantitative data. The aim of the
study is to conduct a quantitative assessment of the contribution to learning through the use of the devel-
oped web-based tool. The research instruments include a controlled experiment to test the efficiency of
the software tool and a survey to gather user feedback on satisfaction. The objective of the survey is an
objective assessment of the tool’s usability.
For the experiment, a purposive (nonprobabilistic) sampling technique was applied (Vehovar et al.,
2016). The selected sample consisted of students who had not yet attended the Machine Learning course
and who had no experience with using the SVM algorithm but had general knowledge about the machine
learning concept. A total of 38 undergraduate students from the Faculty of Technical Sciences in Kosovska
Mitrovica and the Toplica Academy of Vocational Studies in Prokuplje were selected. For the experiment, the
selected sample of students was divided into two groups: control and experimental. The allocation of stu-
dents to the groups was random. Both the experimental and control groups consisted of 19 students each.
Experiment methodology
The experiment consisted of three phases: pre-testing, training, and post-testing (Dugard and Tod-
man, 1995). The pre-test questionnaire was designed with questions related to general information about
classification algorithms and the SVM algorithm. This questionnaire assesses the prior knowledge of all
participants on the given topic. It contained eight multiple-choice quiz questions and six questions requir-
ing standard written answers. Quiz questions were scored with 5 points each, while the standard ques-
tions were scored with 10 points each (or less if the answer was partial).
The post-test questionnaire was similar in structure to the pre-test questionnaire, but the questions
focused solely on the SVM algorithm. The post-test was used to assess the knowledge gained after the
training phase. The training process for the control group involved traditional teaching, while the learning
strategy for the experimental group is described in detail in the Case Study section. Traditional teaching
was also based on the theory-examples-exercises paradigm. For each topic, theoretical concepts were
first presented during lecture sessions. Laboratory exercises were designed to provide a more detailed
www.ijcrsee.com
185
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
explanation of the topics covered in lectures, without the use of an educational web tool, but rather
through practical examples applying the hands-on tasks pedagogical method. The instructors and exam-
iners were the same for both groups.
Survey methodology
In addition to the controlled experiment, it was necessary to conduct a survey on the advantages
and significance of the auxiliary learning system. This process aimed to assess students’ experiences
with the alternative learning method using the auxiliary tool and their perception of the software system’s
effectiveness. For this purpose, the standardized System Usability Scale (SUS) was used. SUS is a quick
and reliable technique for measuring system usability (Brooke, 1996). It consists of a standard question-
naire with 10 questions. For each question, respondents select a statement indicating their level of agree-
ment or disagreement with the statement on a 5-point Likert scale. SUS is generally administered after
the respondent has had the opportunity to use the system being evaluated but before any other testing
or discussion takes place. If a respondent feels unable to answer a particular item, they should mark the
central point on the scale.
Experiment results
For this experiment, quantitative data were collected based on the established methodology, and
descriptive statistics were used for their analysis. To reliably examine the existence of statistically sig-
nificant differences in the achieved results between students in the control and experimental groups, the
ANCOVA statistical method was used. The analyses were conducted using SPSS software (version 25).
ANCOVA (Analysis of Covariance) is used to determine the presence of significant differences between
two or more independent groups concerning a dependent variable. By isolating the effect of a categori-
cal independent variable on the dependent variable, researchers can draw more accurate and reliable
conclusions from their data. ANCOVA seeks differences in adjusted mean values.
Individual differences between students in academic abilities can significantly impact learning out-
comes. Even within the same group, there can be significant variations in students’ prior knowledge.
This variation can obscure the true impact of the applied method on the learning outcomes or results. By
including pre-test results as a covariate in the ANCOVA model, it is possible to more clearly and precisely
understand whether students’ success in the post-test was due to the applied teaching method. In this
statistical model, the post-test result is taken as the dependent variable, the teaching method as the cat-
egorical factor, and the pre-test result as the covariate.
• H
0
- Null Hypothesis: There is no significant difference in the adjusted mean values of the post-test
results between the control and experimental groups.
• H
1
- Alternative Hypothesis: There is a significant difference in the adjusted mean values of the post-
test results between the control and experimental groups.
The achieved post-test results of the students are shown in Figure 12.
Figure 12. Achieved post-test results
www.ijcrsee.com
186
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
Descriptive statistics for the dependent variable (post-test results) are presented in Table 1. Here,
the mean values and standard deviations of the results for each group are shown.
Table 1. Descriptive Statistics for Dependent Variable: Posttest
Group Mean Std. Deviation N
control 69,4737 14,22974 19
experimental 78,6842 12,67567 19
Total 74,0789 14,08734 38
The results of Levene’s Test for Homogeneity of Variances are shown in Table 2. This test evalu-
ates whether the variances between groups are approximately equal.
Table 2. Levene’s Test of Equality of Error Variances
a
Dependent Variable: Posttest
F df1 df2 Sig.
,130 1 36 ,721
a. Design: Intercept + Pretest + Group
The main part of the ANCOVA test results is shown in Table 3. This table presents the effects of the
test between subjects, focusing on the level of significance concerning the independent variable.
Table 3. Tests of Between-Subjects Effects
Dependent Variable: Posttest
Source Type III Sum of Squares df Mean Square F Sig. Partial Eta Squared
Corrected Model 892,054
a
2 446,027 2,420 ,104 ,121
Intercept 3024,956 1 3024,956 16,413 ,000 ,319
Pretest 86,133 1 86,133 ,467 ,499 ,013
Group 868,456 1 868,456 4,712 ,037 ,119
Error 6450,710 35 184,306
Total 215875,000 38
Corrected Total 7342,763 37
a. R Squared = ,121 (Adjusted R Squared = ,071)
A comparison of the mean values of post-test results between the control and experimental groups
is presented in Table 4.
Table 4. Pairwise Comparisons
Dependent Variable: Post
(I) Group (J) Group
Mean Difference (I-J) Std. Error Sig.
b
95% Confidence Interval for Difference
b
Lower Bound Upper Bound
control experimental -9,674
*
4,456 ,037 -18,721 -,627
experimental control 9,674
*
4,456 ,037 ,627 18,721
Based on estimated marginal means
*. The mean difference is significant at the ,05 level.
b. Adjustment for multiple comparisons: Bonferroni.
www.ijcrsee.com
187
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
Survey results
The SUS provides a score that represents a composite measure of the overall usability of the sys-
tem under study. To calculate the score, first sum the contributions from each item. Contributions range
from 0 to 4. For questions 1, 3, 5, 7, and 9, the contribution is the scale position minus 1. For items 2, 4,
6, 8, and 10, the contribution is 5 minus the scale position. The sum of contributions is then multiplied
by 2.5 to obtain the overall score. SUS scores range from 0 to 100. Bangor et al. (2008) experimentally
demonstrate that software with SUS scores below 50 is “unacceptable,” while scores above 70 indicate
“acceptable” usability. Scores between 50 and 70 are considered “marginally acceptable.”
Following the completion of the post-test, students in the experimental group who had the oppor-
tunity to use the educational tool received the SUS questionnaire. The results of this survey, processed
using the described technique, are presented in Table 5.
Table 5. Responses to Individual System Usability Scale Statements
Statement M SD
1 I think that I would like to use this system frequently. 3,47 0,84
2 I found the system unnecessarily complex. 2,63 0,90
3 I thought the system was easy to use. 3,37 0,68
4 I think that I would need the support of a technical person to be able to use this system. 2,89 0,99
5 I found the various functions in this system were well integrated. 3,63 0,68
6 I thought there was too much inconsistency in this system. 2,74 0,81
7 I would imagine that most people would learn to use this system very quickly. 3,53 0,70
8 I found the system very cumbersome to use. 2,58 1,02
9 I felt very confident using the system. 3,42 0,84
10 I needed to learn a lot of things before I could get going with this system. 2,32 1,06
Discussion
The results of the post-test show that the average score of students in the experimental group was
higher than that of the control group. This is clearly seen in Table 1, which displays the descriptive sta-
tistics, revealing an average score of 78.68 for the experimental group compared to 69.47 for the control
group. However, it was crucial to prove that this improvement in the experimental group’s scores was
statistically significant. Therefore, Levene’s test was conducted initially to test for equality of variances.
The results of this test, which are shown in Table 2, with Sig. = 0.721, indicate that there was no signifi-
cant difference in variance between the control and experimental groups because the significance level
exceeded the threshold of 0.05. Thus, based on the pre-test results, it was established that the formed
groups did not statistically differ before the start of the training.
Finally, the results of the ANCOVA analysis (Table 3) demonstrated a statistically significant impact
of using the instructional method with the supplementary software tool on the post-test results. With F (1,
35) = 4.712 and a p-value of 0.037 < 0.05, the null hypothesis (H0) is rejected, supporting the alterna-
tive hypothesis (H1) that there is a significant difference in adjusted mean post-test scores between the
control and experimental groups.
Table 4 compares the mean post-test scores, showing a difference of 9.67% between the mean
scores of the experimental and control groups.
The usability assessment of the system was conducted using the SUS technique, and the results of
this survey are presented in Table 5. This table displays the mean scores for each question. By summing
these mean scores, a total contribution of 30.58 was obtained. As described, this contribution is multiplied
by 2.5 to yield a total SUS score of 76.45. Therefore, since the final score is >70, the implemented tool is
considered acceptable on the SUS usability scale.
Finally, it is important to highlight potential threats to the validity of this empirical assessment.
www.ijcrsee.com
188
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
These threats can be classified as internal, construct, and external validity.
Internal validity threats primarily concern causal questions regarding the presented results. It’s
worth noting that none of the students who participated in the experiment had prior knowledge of the study
preparation. However, if anyone somehow obtained information about the experiment, it could have led
to bias in evaluation. Since collaboration among students is important for the learning process, it cannot
be accurately assessed whether one group of students has better or worse communication and teamwork
skills. To mitigate this threat, both groups had the same instructors.
Construct validity threats usually relate to potential errors in assessment. This type of threat can be
influenced by the evaluation method used. For evaluating this tool, a pretest-posttest method was applied.
Different results may have been obtained using different assessment methods. To reduce the impact of
this threat on the assessment of tool effectiveness, the standard SUS technique, which is widely accepted
for evaluating usability levels, was applied.
External validity threats concern the extent to which findings from the experiment can be general-
ized and how relevant they are to other students beyond this study. To mitigate this threat, students from
two different higher education institutions participated in the experiment. However, for generalization,
one should consider the sample size of students, geographic distance between participating universities,
socio-cultural differences, and the complexity of tests assigned to students. Thus, the results presented in
this study cannot be generalized until more data are obtained from other empirical evaluations involving
different universities and students.
Conclusion
Acquiring skills and knowledge in machine learning is a crucial aspect of academic education for
IT engineers. Teaching complex algorithms in this field should consider the use of innovative tools and
methodologies developed to enhance and supplement the learning and teaching processes. The web
application for learning Support Vector Machines (SVM) represents a significant advancement in com-
puter engineering education, providing intuitive and interactive ways to understand and apply SVM algo-
rithms. Developed using state-of-the-art web technologies, the proposed application promises to become
a valuable resource for students, educators, and researchers in the field of machine learning. This study
analyzes the impact of implementing a new interactive learning method based on the developed web ap-
plication on learning outcomes.
As a result of using the visual software tool in machine learning classes, student engagement levels
have increased. Visualization of the complex SVM algorithm has enabled students to better understand
the workings of the algorithms, down to finer details. These claims are substantiated by a conducted
experiment aimed at quantitatively assessing the learning contributions using the developed web-based
tool. Research instruments included a controlled experiment to verify the efficiency of the software tool
and a survey to collect user satisfaction feedback. The experiment results show that students who used
the tool in their learning (experimental group) have a better understanding of the taught topics compared
to students whose learning was based solely on traditional methods (control group). Through the applica-
tion of the ANCOVA statistical method, the study reliably demonstrated a statistically significant difference
in outcomes between the control and experimental groups.
The usability assessment of the proposed system was evaluated using the SUS technique. Based on the
final survey results, it can be stated that the implemented tool was rated as acceptable on the SUS usability scale.
In future work and research, a broader assessment of the tool with a greater number of students
from various universities should be considered. In addition, future work will be valuable to analyze the
advantages and limitations of the developed web application compared to similar existing tools. Further
enhancements to the proposed system could involve the addition of modules for learning other machine
learning algorithms.
Acknowledgments
The authors would like to express their gratitude to the respondents who participated in the re-
search and the reviewers whose constructive suggestions significantly enhanced the quality of this work.
www.ijcrsee.com
189
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
Author Contributions
Conceptualization, N.J., S.J. and S.S.; methodology, N.J., S.J. and S.S.; software, N.J. and S.J.; formal
analysis, S.S., D.M. and N.S.; writing—original draft preparation, N.J. and S.S.; writing—review and editing,
S.J., D.M. and N.S. All authors have read and agreed to the published version of the manuscript.
Conflict of interests
The authors declare no conflict of interest.
References
Abadi, M., Agarwal, A., Barham, P. et al. (2015). TensorFlow: Large-scale machine learning on heterogeneous systems. https://
doi.org/10.48550/arXiv.1603.04467
Bangor, A., Kortum, P.T. and Miller, J.T., (2008). An empirical evaluation of the system usability scale. Intl. Journal of Human–
Computer Interaction, 24(6), 574-594. https://doi.org/10.1080/10447310802205776
Bennett, K.P. and Campbell, C., (2000). Support vector machines: hype or hallelujah?. ACM SIGKDD explorations newsletter,
2(2), 1-13. https://doi.org/10.1145/380995.380999
Brooke, J. (1996). SUS: A “quick and dirty” usability scale. In P. W. Jordan, B. Thomas, B. A. Weerdmeester, & I. L. McClelland
(Eds.), Usability evaluation in industry, 189–194. London: Taylor & Francis.
Çağlayan, C., (2019). Comparison of the Code-based or Tool-based Teaching of the Machine Learning Algorithm for the
First-Time Learners. In 2019 1st International Informatics and Software Engineering Conference (UBMYK), 1-3. IEEE.
https://doi.org/10.1109/UBMYK48245.2019.8965519
Chang, C. C., & Lin, C. J. (2011). LIBSVM: a library for support vector machines. ACM transactions on intelligent systems and
technology (TIST), 2(3), 1-27. https://doi.org/10.1145/1961189.1961199
Chatzimparmpas, A., Martins, R. M., Jusu, I., & Kerren, A. (2020). A survey of surveys on the use of visualization for interpret-
ing machine learning models. Information Visualization, 19(3), 207-233. https://doi.org/10.1177/1473871620904671
Chollet F., et al. (2015). Keras. https://keras.io
Cortes, C. and Vapnik, V., (1995). Support-vector networks. Machine learning, 20, 273-297. https://doi.org/10.1007/BF00994018
DÍaz, J.M., Dormido, S. and Rivera, D.E., (2015). Interactive Education for Time-Domain Time Series Analysis using ITTSAE.
IFAC-PapersOnLine, 48(28), 751-756. https://doi.org/10.1016/j.ifacol.2015.12.220
Djordjevic, J., Nikolic, B. and Milenkovic, A., (2005). FlexibleWeb-Based Educational System for Teaching Computer Archi-
tecture and Organization. IEEE Transactions on Education, 48(2), 264-273. https://doi.org/10.1109/TE.2004.842918
Dugard, P. and Todman, J., (1995). Analysis of pre‐test‐post‐test control group designs in educational research. Educational
Psychology, 15(2), 181-198. https://doi.org/10.1080/0144341950150207
Guo, P., Saab, N., Post, L.S. and Admiraal, W., 2020. A review of project-based learning in higher education: Student outcomes
and measures. International journal of educational research, 102, 101586. https://doi.org/10.1016/j.ijer.2020.101586
Hall, M., Frank, E., Holmes, G., Pfahringer, B., Reutemann, P., & Witten, I. H. (2009). The WEKA data mining software: an
update. ACM SIGKDD explorations newsletter, 11(1), 10-18. https://doi.org/10.1145/1656274.1656278
Hazzan, O., and Mike, K. (2023). Guide to Teaching Data Science: An Interdisciplinary Approach. Springer Nature.
Jeong, D. H., Ziemkiewicz, C., Fisher, B., Ribarsky, W., & Chang, R. (2009). iPCA: An interactive system for pca‐based visual
analytics. In Computer Graphics Forum (Vol. 28, No. 3, pp. 767-774). Oxford, UK: Blackwell Publishing Ltd. https://doi.
org/10.1111/j.1467-8659.2009.01475.x
Jovanović, N., Popović, R., Marković, S. and Jovanovic, Z., (2012). Web laboratory for computer network. Computer Applica-
tions in Engineering Education, 20(3), 493-502. https://doi.org/10.1002/cae.20417
Jovanović, N., Stamenković, S., and Jovanović, S. (2023). NNeduca: a software environment to teach articial neural net-
works, Comput. Appl. Eng. Educ. Volume 31, Issue 5, 1447–1464. https://doi.org/10.1002/cae.22655
Kim, J. T., Kim, S., & Petersen, B. K. (2021). An interactive visualization platform for deep symbolic regression. In Proceedings
of the Twenty-Ninth International Conference on International Joint Conferences on Articial Intelligence, 5261-5263.
https://doi.org/10.24963/ijcai.2020/763
Mayeld, E., & Rosé, C. (2010). An interactive tool for supporting error analysis for text mining. In Proceedings of the NAACL
HLT 2010 Demonstration Session, 25-28.
Mühlbacher, T., Piringer, H., Gratzl, S., Sedlmair, M. and Streit, M. (2014). Opening the black box: Strategies for increased user
involvement in existing algorithm implementations. IEEE transactions on visualization and computer graphics, 20(12),
1643-1652. https://doi.org/10.1109/TVCG.2014.2346578
Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L. and Desmai-
son, A., (2019). Pytorch: An imperative style, high-performance deep learning library. Advances in neural information
www.ijcrsee.com
190
Jovanović, N. et al. (2025). A Web Application for Learning Support Vector Machine Algorithms in Computer Engineering,
International Journal of Cognitive Research in Science, Engineering and Education (IJCRSEE), 13(1), 175-190.
processing systems, 32. https://doi.org/10.48550/arXiv.1912.01703
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg,
V. and Vanderplas, J., (2011). Scikit-learn: Machine learning in Python. The Journal of machine Learning research, 12,
2825-2830. https://www.jmlr.org/papers/volume12/pedregosa11a/pedregosa11a.pdf?source=post_page
Pradono, S., Astriani, M.S. and Moniaga, J., (2013). A method for interactive learning. International Journal of Communication
& Information Technology, 7(2), 46-48. https://doi.org/10.21512/commit.v7i2.583
Rutten, N., Van Joolingen, W. R., and Van der Veen, J. T., (2012). The learning effects of computer simulations in science
education. Computers & Education, Elsevier, 58(2012), 136-153. https://doi.org/10.1016/j.compedu.2011.07.017
Sacha, D., Sedlmair, M., Zhang, L., Lee, J.A., Peltonen, J., Weiskopf, D., North, S.C. and Keim, D.A. (2017). What you see is
what you can change: Human-centered machine learning by interactive visualization. Neurocomputing, 268, 164-175.
https://doi.org/10.1016/j.neucom.2017.01.105
Schölkopf, B. and Smola, A.J., (2002). Learning with kernels: support vector machines, regularization, optimization, and be-
yond. MIT press.
Silva, W., Steinmacher, I. and Conte, T. (2019). Students’ and instructors’ perceptions of ve different active learning strategies
used to teach software modeling. IEEE Access, 7, 184063-184077. https://doi.org/10.1109/ACCESS.2019.2929507
Sonnenburg, S., Rätsch, G., Henschel, S., Widmer, C., Behr, J., Zien, A., Bona, F.D., Binder, A., Gehl, C. and Franc, V.,
(2010). The SHOGUN machine learning toolbox. The Journal of Machine Learning Research, 11, 1799-1802. https://
doi.org/10.5446/19980
Stamenković, S., Jovanović, N. (2024). A Web-based Educational System for Teaching Compilers, IEEE Transactions on
Learning Technologies. Vol. 17, 143-156. https://doi.org/10.1109/TLT.2023.3297626
Stamenković, S., Jovanović, N., Vasović, B., Cvjetković, M. and Jovanović, Z., (2023). Software tools for learning articial intel-
ligence algorithms. Articial Intelligence Review, 1-30. https://doi.org/10.1007/s10462-023-10436-0
Taher, M., and Khan, A., (2014). Impact of Simulation-based and Hands-on Teaching Methodologies on Students’ Learning
in an Engineering Technology Program. Proceedings of the ASEE Annual Conference & Exposition, 1-22. https://doi.
org/10.18260/1-2--20593
Thakur, J., and Kumar, N. (2011). DES, AES and Blowsh: Symmetric key cryptography algorithms simulation based perfor-
mance analysis. International journal of emerging technology and advanced engineering, 1(2), 6-12.
Vehovar, V., Toepoel, V. and Steinmetz, S., (2016). Non-probability sampling. The Sage handbook of survey methods. https://
doi.org/10.4135/9781473957893.n22
Zeiler, M.D. and Fergus, R. (2014). Visualizing and understanding convolutional networks. In Computer Vision–ECCV 2014:
13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part I 13, 818-833. Springer
International Publishing. https://doi.org/10.48550/arXiv.1311.2901
