Methods, Models and Techniques to Improve Information System’s
Security in Large Organizations
Vladislavs Minkevics and Janis Kampars
Riga Technical University, Kalku 1, Riga, Latvia
Keywords: IS Security, Big Data, Malware, Security Methods and Techniques, DGA.
Abstract: This paper presents the architecture of a modular, big-data based IS security management system (ISMS) and
elaborates one of its modules – the domain generation algorithm (DGA) generated domain detection module.
The presented methods, models and techniques are used in Riga Technical University, and can be used in any
other large organization to stand against IS security challenges. The paper describes how organization can
construct IS security management system using mostly free and open source tools and reach it’s IS security
goals by preventing or minimizing consequences of malware with little impact on employee’s privacy. The
presented DGA detection module provides detection of malicious DNS requests by extracting features from
domain names and feeding them into random forest classifier. ISMS doesn’t rely solely of DGA detection
and instead uses an ensemble of modules and algorithms for increasing the accuracy of the overall system.
The presented IS security management system can be employed in real-time environment and its DGA
detection module allows to identify infected device as soon as it starts to communicate with the botnet
command and control centre to obtain new commands. The presented model has been validated in the
production environment and has identified infected devices which were not detected by antivirus software nor
by firewall or Intrusion Detection System.
1 INTRODUCTION
In our digital society, where every person relies on
Internet in one or another way, securing information
systems have become a challenge like never before.
For security reasons institutions like banks,
healthcare, insurance organisations must meet certain
security standards for example PCI DSS, ISO27001
("Information technology-Security techniques-Code
of practice for information security controls"). These
standards define IT security’s technical and
organisational measures to ensure minimization of IT
security risks on data confidentiality, integrity and
availability. Nevertheless many reports have surfaced
of IT security breaches in companies that complied
with the appropriate certifications (The 18 biggest
data breaches of the 21st century, 2018) (PCI DSS:
Lessons to learn from recent payment card breaches,
2018). Traditional protection methods alone, like
firewalls, IDS/IPS systems, are no longer adequate to
deal with ever growing number of threats.
According to Trustwave 2019 security report
(Trustwave, 2019), one of the major problems in IT
security is the substantial amount of time that elapses
from penetration until the breach is finally identified.
The report shows that median number of latencies in
days for internally detected incidents is 11 days. The
other critical problem in large organizations is high
number of vulnerabilities, especially if different
operating systems are used. According to this report
100% of globally scanned resources contained at least
one vulnerability and 9% of discovered
vulnerabilities were high risk or critical.
To address these risks this paper proposes a
modular, Big Data based IS security management
system (ISMS). The goal of this paper is to define the
architecture of ISMS and elaborate one of its modules
the domain generation algorithm (DGA) detection
module, by identifying the best DGA domain name
detection features and machine learning algorithm.
One of the distinguishing features of this research is
use of real-time data from production environment in
Riga Technical university (RTU) and the ability to
check for true and false positives. Our DNS data
originates from devices used by RTU students,
employees and guest researchers. This provides more
realistic evaluation of machine learning based DGA
domain name detection module if compared with
existing studies that use publicly available datasets.
632
Minkevics, V. and Kampars, J.
Methods, Models and Techniques to Improve Information System’s Security in Large Organizations.
DOI: 10.5220/0009572406320639
In Proceedings of the 22nd International Conference on Enterprise Information Systems (ICEIS 2020) - Volume 1, pages 632-639
ISBN: 978-989-758-423-7
Copyright
c
2020 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
DGA detection is necessary because currently
modern malware is trying to be as stealthy as
possible. Especially this applies to botnets. DGA was
introduced by botnets, so that the domain name is
continuously changing and is difficult to block.
Infected device always knows which domain it should
resolve to reach the botnet’s command and control
centre. Most of these algorithmically generated
domains are very awkward to human eye.
This work is structured as follows. Section 2
provides a review of the related work in the field of
IS security management and DGA detection. Section
3 presents the IS security management background in
Riga Technical University and further motivates the
development of ISMS. Section 4 defines the
architecture of the ISMS. Section 5 develops and
evaluates the DGA detection module in production
environment. Section 6 concludes and provides
directions for future research.
2 RELATED WORK
To fight with IS security challenges unified security
systems must be used. Alguliyev et al (Alguliyev &
Imamverdiyev, 2014) suggest to use Big Data for IT
security, addressing challenges like advanced
persistent threats, detections of data leakage, incorpo-
ration of forensics, fraud and criminal intelligence.
Nevertheless, they list several challenges in the Big
Data field at time of writing: privacy, lack of Big Data
based detection algorithms, security visualization
problems and lack of skilled personnel. Some of these
challenges, like Big Data based detection algorithms
have already been addressed today.
There are several studies concentrating on the
DGA detection problem. Researchers have defined
the most important features that can be used to train
machine learning algorithms for DGA domain name
identification. Feature selection is covered by
(Barbosa, Souto, Feitosa, & El-Khatib, 2015)
(Jonathan Woodbridge, Hyrum S. Anderson, Anjum
Ahuja, 2016) (Jose Selvi, Ricardo J.Rodríguez,
2019). Not all of the proposed DNS features and
resulting models contribute to better detection of
DGA domain names.
Mowbray et al. (Mowbray & Hagen, 2014)
suggest that unusual distribution of second-level
string lengths in the domain name is a good indicator
for it being a DGA generated domain name. The
proposed algorithm looks for domains with an
unusual distribution of lengths, however it can only
detect DGAs that are used for second-level domain
fluxing. Once domains from a new malware DGAs
have been detected, they use it to retrain a classifier
so that the new DGAs can be detected with a better
accuracy.
Truong et al. (Truong & Cheng, 2016) propose to
use decision trees to train a DGA classifier. The
algorithm reports accuracy of 92.3%. The research
suggests using length as one of the features for DGA
detection. Authors conclude that DGA detection with
machine learning algorithms should not be used as the
only solution for botnet detection, since new
generation botnets tend to use a technique called
domain fluxing.
A research by Ahluwalia et al. (Ahluwalia,
Traore, Ganame, & Agarwal, 2017) shows that
domain length plays an important role in DGA
detection and has great impact on the detection
accuracy. Decision trees based DGA generated
domain name detection model accuracy for 6-
character long domain names is 85,15%, while it
reaches 94% for 10-character long domain names.
Authors use n-grams to construct features. The
experimental results show that random forests have
slightly better performance in DGA generated
domain name detection.
Selvi et al. (Jose Selvi, Ricardo J.Rodríguez,
2019) analyse 32,000 malware domains and propose
to use a set of lexical features based on masked n-
grams as features detecting DGA generated domains.
The research distinguishes 18 features and concludes
that a combination of lexical and statistical features
together with bigrams shows the best results in terms
of accuracy. The performed experiments indicate that
the most important features are unigrams, bigrams
and trigrams, their standard deviation as well as
number of consonants divided by domain name
length. Three different machine learning models are
examined: nearest neighbours, decision trees, and
random forests, while the latter achieves the best
accuracy.
Majority of the previously done studies in this
area relies on publicly available datasets, therefore
due to poor generalization models might not be
efficient if used in the production environment.
J. Peck et al. (Peck, u.c., 2019) suggest that using
machine learning algorithm as the only means for
detecting DGA generated domain names is not
sufficient and other IS security management methods
should also be employed. Another reason for using
additional approaches for malware detection is
masking of DNS traffic using DNS over TLS (Z.
Hu,L. Zhu,J. Heidemann,A. Mankin,D. Wessels,
2016) and DNS over HTTPS (Mozilla.org, 2020).
Many companies like Cloudflare, Quad9, Google
have already announced public DNS resolver services
Methods, Models and Techniques to Improve Information System’s Security in Large Organizations
633
via DNS over TLS (Quad9, 2020) (Cloudflare, 2020)
(Google, 2019).
There are studies concentrating on the IP Flow
analysis problem in real time. IP Flow data is valuable
because it provides information about connections,
TCP/IP flags and data amount sent and received. This
data can be used to detect suspicious activity and
compromised devices. Jirsik et al. (Jirsik, Cermak,
Tovarnak, & Celeda, 2017) claim that IP Flow data
aids traditional monitoring with the ability to run
analytical queries that are evaluated in real time with
high throughput, low latency, and good scalability, all
at the same time and allows security analysts to
perform real-time analysis on network data and detect
network attacks instantly, and provides them with a
deep understanding of the network via in-depth
situational awareness.
3 BACKGROUND
The development of first generation IS security
management platform (ISSMP) in RTU started in
2014. Initially it consisted of Suricata IDS (Suricata,
2020) and Python scripts that helped chief security
officer to analyse IDS generated data.
Platform relied on open source products and the
main automation was based on cronjob script
execution. Python scripts were used to cover all
phases of the automated threat detection logic. The
system sent notifications to specific user whose
device was classified as infected and provided
additional details to Chief security officer.
Platform allowed RTU to dramatically decrease
the number of notifications by Information
Technology Security Incident Response Institution of
the Republic of Latvia (CERT) regarding
compromised IPs (see Fig.1.).
The ability to automatically block IPs in firewall,
which was introduced in 2019, enabled further
reduction of incidents, since it was very effective
measure in scenarios where the notified user is not
taking action to clean his/her device from malware.
Figure 1: Compromised IPs in RTU by Latvian CERT.
Despite the achieved progress there is still room
for improvement, considering that not all infections
are detected by CERT and amount of data for
processing increased. Several issues have been
discovered and ISSMP was unable to address them
due to various architectural limitations. The amount
of processable data in some of system’s components
cannot be handled by standalone Python scripts. The
other discovered problem is the latency of data
processing, which needs to be minimized. To address
these challenges Big Data concept was adopted and a
new generation of IS security management platform
(ISMS) is presented in Section 4.
4 ISMS ARCHITECTURE
The next generation of IS security management
platform (ISMS) is being developed at RTU. The
system is based on well proven Big Data technologies
and it will incrementally replace the previously used
system.
The ISMS has both preventive and detective
capabilities. It contains vulnerability management
component which provides regular manual scans and
preparation of analytical reports for the responsible
personnel. Automation of vulnerability identification
process is used to:
1) Identify hosts that should be scanned, for
example using specific open port. For this
purpose, Nmap (Nmap, 2020) is used;
2) Normalize the results from Nmap scan and
feed IP addresses into Nessus vulnerability
scanner (Tenable, 2020);
3) Scan IP addresses for vulnerabilities;
4) Parse the result of Nessus scanner and extract
Critical vulnerabilities with corresponding
IP’s;
5) Sort the results according to responsible
personnel and send an email with IP addresses
and vulnerabilities that their devices have.
The architecture of the proposed ISMS is based on
capability driven development (Sandkuhl & Stirna,
2018) which allows better traceability of
organisational IS security management goals and
their fulfilment (Minkevics & Kampars, 2018). ISMS
contains open source products and is capable of
handling security incidents in an efficient manner
with respect to privacy according to EU General data
protection regulation (EU, 2016) consists of two
major and four minor parts (see Fig.2.).
ICEIS 2020 - 22nd International Conference on Enterprise Information Systems
634
The Big Data paradigm is used in the ISMS to
address the following challenges:
1) data are very large and cannot be processed
by traditional methods;
2) data are produced with great velocity and
must be captured and processed rapidly;
3) data have different structure. (Alguliyev &
Imamverdiyev, 2014).
These challenges can be addressed with parallel
processing and tools built to handle Big Data like
Hadoop, Apache Spark, Cassandra, Apache Kafka.
In Big Data context, an open source, distributed
platform Apache Kafka (Kafka, 2020) which handles
messages in real time was adopted. Introducing Kafka
into the automated security management process in
RTU has allowed to decrease the reaction time by up
to 10 seconds (previously it took up to 80 seconds to
react on malicious activity). Further improvement of
the ISMS will provide even greater reduction of the
latency.
The privacy goals in the ISMS are reached by not
linking the IP address to a user prior to registered
suspicious activity. While constant IP address user
lookups for all users can speed up the process of
incident analysis, it has negative effect on the user
privacy. Authors believe that the potential delay of a
few seconds for IP user lookup when malicious
activity is identified is neglectable and the value of
improved privacy is much greater, especially if DNS
analysis is in place.
The ISMS provides the following means of
automation:
1) Automated Suricata IDS (Suricata IDS , 2020)
rule update and message transfer to Kafka;
2) Automated message forwarding from Firewall
to Kafka;
3) Automated DNS request transfer to Kafka;
4) Automated message retrieval from Kafka and
further analysis:
a. Checking if known malware DNS requests
(Malware domain list, 2020) are in DNS
data;
b. Checking if DGA generated domain name
detection module has identified malicious
DNS requests (see Section 5);
c. Checking if the generated alert is not false
positive;
5) Notifying user and Chief security officer.
The ISMS consists of analysis and actions
modules, where the analysis module consists of many
sub-modules for vulnerability identification, open or
vulnerable service identification, detecting malicious
behavior by a device or user. The user activity sub
module consists of portal login information analysis,
where information from successful portal logins are
collected and analyzed to identify if user has logged
in from other country and home country within last
24 hours. A similar analysis is performed for Office
365 user logins. DHCP lease logs are collected to
obtain information about device, so change of IP
address does not impact accuracy of device
identification. Suricata IDS and Firewall logs and
DNS traffic data are collected for detecting infected
devices. Switch logs are collected to obtain
information about the device’s physical location if it
is part of the local network. IP flow data is obtained
and sent to Apache Kafka. At the moment IP Flow
data is used to manually analyze and confirm
suspicious IP address activity. During the analysis the
full picture of connections and DNS requests is
created for time period of possible malicious activity.
Availability of this information supports well-
motivated data-driven decisions by the Chief security
officer. The traffic data module is used when brute
force is detected and collects full network data of a
suspected device for the period of one minute.
Figure 2: Architecture of ISMS system.
The action component consists of different
notification modules which provide user notification
via SMS, e-mail and internal portal. The access denial
module interacts with firewall API to deny access for
specific device in case user is not taking action to stop
the detected malicious activity.
Methods, Models and Techniques to Improve Information System’s Security in Large Organizations
635
5 DGA DETECTION MODULE
IMPLEMENTATION AND
EVALUATION
The following steps were performed to develop the
DGA generated domain name detection model:
1) preparation of training dataset,
2) feature and classifier selection based on
related work research and experiments,
3) performance evaluation.
The training data was collected for 3 months (from
04.10.2019-04.01.2020) by recording actual DNS
resolution requests in RTU production environment.
To enable training of the classifier, the acquired
domain names needed to be classified into legitimate
and malicious domains. This was done by following
the logic described in Table 1 (rule-based automated
semantic analysis and information retrieval from
external sources like virustotal.com and quad9.com)
and Table 2 (semi-automated classification of DNS
record based on output from Table 1). Prior to that,
the gathered data was cleaned from DNS requests for
non-existing domains (the requested domain name
contains a spelling error). This was done by checking
whether the requested domain names is in icann.org
root zone database (ICANN root zone, 2020).
Table 1: Specific ruleset No.1. dataset creation.
Rule
No.
Description
Examples
1
IF DOMAIN CONTAINS
NO WOWELS THEN
SUSPICIOUS++
0sntp7dnrr.com,
pnghst.com
2
IF DOMAIN CONTAINS
NO CONSONANTS
THEN SUSPICIOUS++
3458ee.com, o5o4o6.com
3
IF DOMAIN CONTAINS
ONLY NUMBERS THEN
SUSPICIOUS++
127777.com,
10000114.com,
12688888.com
4
IF DOMAIN CONTAINS
ONLY HEXIDECIMAL
THEN SUSPICIOUS++
442d9f2ac50ca502.com,
8cb0309458c7b35e.com
5
IF DOMAIN CONTAINS 3
WOWELS IN A ROW
THEN SUSPICIOUS++
zwyr157wwiu6eior.com,
zy16eoat1w.com
6
IF DOMAIN CONTAINS 3
CONSONANTS IN A ROW
THEN SUSPICIOUS++
yqezqofkb1nnmz.com,
6l1twlw9fy.com
7
IF DOMAIN CONTAINS 5
WOWELS IN A ROW
THEN SUSPICIOUS++
booooooom.com,
iiiiiiiiiiii.net
8
IF DOMAIN CONTAINS 5
CONSONANTS IN A ROW
THEN SUSPICIOUS++
yqezqofkb1nnmz.com,
eclkmpbn.com
9
IF DOMAIN CONTAINS
NUMBERS THAT ARE
NOT IN A ROW THEN
SUSPICIOUS++
eh8jq4cmq8j9g5.com,
5kv261gjmq04c9.com
Table 2: Specific ruleset No.2. dataset creation.
Rule
No.
Pseudo Code of specific rule used
1
IF SUSPICIOUS == 0 THEN
IF = EXPERT IDENTIFIED IT AS MALICIOUS
DOMAIN_MALICIOUS = 1; EXIT
ELSE DOMAIN_MALICIOUS = 0; EXIT
ELSE GOTO RULE2
2
IF DOMAIN IN ALEXA TOP 100000 THEN
DOMAIN_MALICIOUS = 0; EXIT
ELSE GOTO RULE3
3
IF DOMAIN IN QUAD9 AS MALICIOUS THEN
DOMAIN_MALICIOUS = 1; EXIT
ELSE DOMAIN_MALICIOUS = 0; EXIT
4
IF DOMAIN IN VIRUSTOTAL AS MALICIOUS
THEN
DOMAIN_MALICIOUS = 1; EXIT
ELSE DOMAIN_MALICIOUS = 0; EXIT
Additionally, top 30000 domains from alexa.com
were added to the training dataset. The final dataset
consisted of 67749 DNS records, where 65999
records were marked as legitimate and 1750 were
marked as malicious. The following experiments
were conducted after training dataset preparation:
Experiment 1 -
selecting the most appropriate
classifier algorithm for DGA. Algorithm is selected
based on existing studies and practical experiments
with selected training set. The initial list of features
were chosen based on research by Selvi et al. and are
summarized in Table 3.
Table 3: Features used to detect DGA.
Feature
No.
Description
Example for
7hu8e1u001.com
F1 length 10
F2 Unigram average 28.39
F3 Bigram average 0.257
F4 Trigram average 0.229
F5
Unigram standard
deviation 0.0
F6
Bigram standard
deviation 3.459
F7
Trigram standard
deviation 3.430
F8 Vowels to length 0.300
F9 Consonants to length 0.100
F10 Unique chars to length 0.001
Due to the fact that natural distribution between
legitimate and malicious DNS requests could result in
a model that has poor DGA detection capabilities,
experiments were performed to determine the optimal
training dataset distribution between legitime and
malicious domain names. We chose four classifiers:
Random forest (RFC), Decision trees (DTC), Neural
networks (NNC), and C-support vector classification
(SVM). To get more objective results, we split our
ICEIS 2020 - 22nd International Conference on Enterprise Information Systems
636
dataset into 18 datasets. We used python
train_test_split library for splitting. Only legitimate
domains were split. Malicious domains were the same
in every dataset. For every training set 10fold cross
validation has been performed using sklearn python
library.
Classificators’ performance for every iteration is
shown in Figures 3-6. Y-axis shows precision, recall,
F1 score and accuracy percentage values, where’s x-
axis corresponds to number of legitimate domains in
the specific dataset.
Figure 3: RFC performance.
Figure 4: DTC performance.
Figure 5: NNC performance.
Figure 6: SVM performance.
The performance was evaluated by getting the mean
values of every iteration’s precision, recall, F1 score
and accuracy (see Table 4).
Table 4: Performance measures of experiment 1.
- Precision Recall F1-Score Accuracy
RFC 0.883 0.822 0.851 0.959
DTC 0.816 0.823 0.819 0.948
NNC 0.831 0.782 0.804 0.949
SVM 0.866 0.753 0.798 0.951
Results of Experiment 1. Our results, similarly to
Ahluwalia et al (Ahluwalia, Traore, Ganame, &
Agarwal, 2017) proved that better performing
machine learning algorithm for our task is Random
forest classifier (RFC). Random forest classifier is
further used to identify DGA in real-time
environment.
Experiment 2 identification of feature set where
our chosen classifier performs best. Specific feature
sets were selected (see Table 5).
Table 5: Feature sets.
- SET1 SET2 SET3 SET4 SET5
F1 111 0 1
F2 010 0 1
F3 111 1 1
F4 111 1 1
F5 010 0 1
F6 111 1 1
F7 111 1 1
F8 111 1 1
F9 111 1 1
F10 001 0 1
The performance was measured using 10fold cross
validation for every feature set and getting the mean
values of precision, recall, F1 score and accuracy.
Additionally, 2 datasets were created with 15
Methods, Models and Techniques to Improve Information System’s Security in Large Organizations
637
malicious and 15 legitimate domain names that were
not in dataset. This was done to evaluate how many
false positives will every classifier produce. The false
positives were calculated by adding incorrectly
identified domains of every iteration. The result is
shown in Table 6.
Table 6: Performance measures of experiment 2.
- Malw
are
FP
Legit
FP
Preci
sion
Recal
l
F1-
Score
Accura
cy
SET1 54 0 0.881 0.825 0.851 0.959
SET2 55 0 0.877 0.823 0.848 0.958
SET3 50 0 0.880 0.816 0.846 0.957
SET4 59 0 0.878 0.820 0.847 0.957
SET5 47 0 0.883 0.822 0.851 0.959
Results of Experiment 2. Experiments showed that
(see Table 6) the best performance of RFC classifier
is achieved by using all 10 features, which is why all
features were also used by the automated DGA
discovery module. It was implemented in production
real-time environment using Apache Spark. DNS
traffic is being aggregated in 5-minute windows.
The RTU production environment tests using the
proposed DGA discovery module were performed
from December 24, 2019 until December 29, 2019.
Table 7: Practical results of implemented DGA detection
module.
DGA detection module performance No.
Detected as malware DNS, and not in training
database 865
Virustotal.com detected DNS as malicious 53
Virustotal.com detected DNS as suspicious 33
Quad9.com detected DNS as malicious 75
Detected as malware DNS, and not in training
database 865
The results (see Table 7) show that the DGA
detection module can be successfully adopted in real
time environment using Big Data concept. Since the
beginning of experiments, more than 30 DGA
detected infection cases (mostly student devices)
were confirmed by the end-device owner using other
antimalware systems.
6 CONCLUSIONS AND FUTURE
WORK
Methods and techniques described in this paper were
adopted by RTU and enabled to address IT security
challenges every organisation faces nowadays. Big
Data is a valuable source for improving overall IT
security in organization. Use of Big Data concepts are
a must for handling large amounts of data in real time,
as it is in cybersecurity. New opportunities arise to
use IP flow data with machine learning algorithms
and improve threat identification even more.
By using the ISMS:
1) The overall rate of false positives dropped;
2) Malware identification rate increased;
3) Malware unknown by security vendors have
been discovered;
4) Reaction speed on incidents raised, and now it
is approximately 10 seconds;
5) We gained ability to scale and extend the
solution as necessary.
The adoption of the new system has stepped up RTU
to real time automated IS security risk management
process adoption. More modules are being developed
and integrated in the platform to increase the accuracy
of threat detection. Experiments with using machine
learning in other areas, like IP flow data analysis are
being performed. The level of the ISMS
autonomousity will be raised by providing better
integration with systems like corporate firewalls,
which will enable automatic blocking of infected
devices. The platform, its methods and techniques can
be adapted by any large organisation facing similar
challenges.
ISMS has already proven its effectiveness, by
detecting approximately 40 new malware infections
per month, which would remain undetected by the
previous generation of IS security management
system. The DGA detection module alone has
allowed to identify more than 30 infections. In future
we will continue to expand ISMS system by adding
machine learning based IP flow data analysis module.
REFERENCES
"Information technology-Security techniques-Code of
practice for information security controls". (n.d.).
27002:2015, ISO/IEC.
Ahluwalia, A., Traore, I., Ganame, K., & Agarwal, N.
(2017). Detecting Broad Length Algorithmically
Generated Domains. In I. Traore, I. Woungang, & A.
Awad (Ed.), Intelligent, Secure, and Dependable
Systems in Distributed and Cloud Environments (pp.
19-34). Cham: Springer International Publishing.
Alguliyev, R., & Imamverdiyev, Y. (2014). Big Data: Big
Promises for Information Security. 2014 IEEE 8th
International Conference on Application of Information
and Communication Technologies (AICT), (pp. 1-4).
Barbosa, K., Souto, E., Feitosa, E., & El-Khatib, K. (2015).
Identifying and Classifying Suspicious Network
ICEIS 2020 - 22nd International Conference on Enterprise Information Systems
638
Behavior Using Passive DNS Analysis. 2015 IEEE
International Conference on Computer and
Information Technology; Ubiquitous Computing and
Communications; Dependable, Autonomic and Secure
Computing; Pervasive Intelligence and Computing,
(pp. 160-167).
Cloudflare. (2020, 01 28). DNS over TLS. Retrieved from
https://developers.cloudflare.com/1.1.1.1/dns-over-tls/
EU. (2016, 04 27). Regulation (EU) 2016/679 of the
European Parliament and of the Council of 27 April
2016 on the protection of natural persons with regard
to the processing of personal data and on the free
movement of such data, and repealing Directive
95/46/EC (General Da. Retrieved 10 24, 2018, from
https://eur-lex.europa.eu/legal-
content/EN/TXT/?qid=1532348683434&uri=CELEX:
02016R0679-20160504
Google. (2019). Google Public DNS now supports DNS-
over-TLS. Retrieved from https://security.google
blog.com/2019/01/google-public-dns-now-supports-
dns-over.html
ICANN root zone. (2020, 01 28). (ICANN.org) Retrieved
01 17, 2020, from http://stats.research.icann.org/dns/
tld_report/archive/index.html
Jirsik, T., Cermak, M., Tovarnak, D., & Celeda, P. (2017).
Toward Stream-Based IP Flow Analysis. IEEE
Communications Magazine, 55(7), 70-76.
Jonathan Woodbridge, Hyrum S. Anderson, Anjum Ahuja,
D. (2016). Predicting Domain Generation Algorithms
with Long Short-Term Memory Networks. 13.
Arlington,VA 22201: Endgame, Inc.
Jose Selvi, Ricardo J.Rodríguez, E.-O. (2019). Detection of
algorithmically generated malicious domain names
using masked N-grams. Elsevier, Volume 124, Pages
156-163.
Kafka, A. (2020, 01 28). (Apache Kafka) Retrieved 08 10,
2019, from https://kafka.apache.org/
Malware domain list. (2020, 01 28). Retrieved 08 10, 2019,
from http://www.malwaredomainlist.com
Minkevics, V., & Kampars, J. (2018). IS Security
Governance Capability Design for Higher Education
Organization. 2018 59th International Scientific
Conference on Information Technology and
Management Science of Riga Technical University
(ITMS), (pp. 1-5).
Mowbray, M., & Hagen, J. (2014). Finding Domain-
Generation Algorithms by Looking at Length
Distribution. 2014 IEEE International Symposium on
Software Reliability Engineering Workshops, (pp. 395-
400).
Mozilla.org. (2020, 01 07). About DNS-over-HTTPS.
(Mozilla.org) Retrieved 01 07, 2020, from
https://support.mozilla.org/en-US/kb/firefox-dns-over-
https#w_about-dns-over-https
Nmap. (2020, 01 28). (NMAP.ORG) Retrieved 08 20, 2019,
from https://nmap.org
PCI DSS: Lessons to learn from recent payment card
breaches. (2018, 09 14). (itgovernance) Retrieved 01
31, 2020, from https://www.itgovernance.co.uk/
blog/pci-dss-lessons-to-learn-from-recent-payment-
card-breaches
Peck, J., Nie, C., Sivaguru, R., Grumer, C., Olumofin, F.,
Yu, B., . . . Cock, M. (2019). CharBot: A Simple and
Effective Method for Evading DGA Classifiers. IEEE
Access, 7, 91759-91771.
Quad9. (2020, 01 28). DoH with Quad9 DNS Servers
.
Retrieved from https://www.quad9.net/doh-quad9-dns-
servers/
Sandkuhl, K., & Stirna, J. (2018). Capability Management
in Digital Enterprises. Springer International
Publishing.
Suricata. (2020, 01 28). Suricata. Retrieved 01 22, 2020,
from https://suricata-ids.org/
Suricata IDS . (2020, 01 28). (The Open Information
Security Foundation.) Retrieved 08 20, 2019, from
https://suricata-ids.org/
Tenable. (2020, 01 28). (Tenable) Retrieved 08 10, 2019,
from https://www.tenable.com/products/nessus/nessus-
professional
The 18 biggest data breaches of the 21st century. (2018, 12
20). (CSO) Retrieved 01 31, 2020, from
https://www.csoonline.com/article/2130877/the-
biggest-data-breaches-of-the-21st-century.html
Truong, D.-T., & Cheng, G. (2016, 9 25). Detecting
domain-flux botnet based on DNS traffic features in
managed network. Security and Communication
Networks, 9(14), 2338-2347.
Trustwave. (2019). 2019 Trustwave Global Security
Report. (Trustwave) Retrieved 08 20, 2019, from
https://www.trustwave.com/en-us/resources/library/
documents/2019-trustwave-global-security-report/
Z. Hu,L. Zhu,J. Heidemann,A. Mankin,D. Wessels, P.
(2016). Specification for DNS over Transport Layer
Security (TLS). 18. IETF Tools.
Methods, Models and Techniques to Improve Information System’s Security in Large Organizations
639