Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR127

Fraud Detection in NoSQL Database Systems

using Advanced Machine Learning
Tamilselvan Arjunan
Lead Software Engineer

Abstract:- NoSQL databases such as MongoDB and This paper presents a comprehensive review of the most
Cassandra have been rapidly adopted in recent years recent machine learning algorithms that can be used to enhance
because of their high performance, flexibility, and intrusion detection and fraud detection within NoSQL
scalability. These databases present new security issues databases. We examine the application of supervised, online,
compared to SQL databases. NoSQL databases are and unsupervised learning models, including neural networks,
vulnerable to fraud, intrusions and data breaches due to classification, clustering and ensemble methods. Research
their dynamic schemas, lack of control over access and the contributions include a classification of NoSQL attacks, feature
focus on availability. This paper examines how advanced engineering techniques to pre-process database telemetry and
machine-learning techniques can be used to enhance fraud novel applications of online learning to adaptive threat
and intrusion detection in NoSQL databases. We examine detection.
different machine-learning algorithms, including neural
networks and support vector machines. Random forests, The remainder of the document is organized as follows.
clustering, and random forests can be used to analyze large The second section provides background information on
databases activity logs in order to identify anomalous NoSQL database security issues. Section 3 examines machine
patterns of access indicative of malicious behavior. We learning techniques used in intrusion detection systems. The
examine how these models are trained online to detect taxonomy for NoSQL-based attacks is presented in Section 4.
emerging threats, and we validate the techniques using Section 5 presents experiments using machine learning
proof-of concept experiments on a prototype NoSQL based algorithms for NoSQL intrusion detection and fraud detection.
database. Our results show high accuracy for detecting The results are analyzed in Section 6. The section 7 concludes
injection attacks, unauthorized query, and abnormal by making recommendations for future research directions.
database traffic, with low false-positive rates.
A. Background
Keywords:- Nosql, Mongodb, Security, Intrusion Detection, NoSQL ("Not Only SQL") databases have risen in
Fraud Detection, Machine Learning. popularity as web-scale applications driven by big data have
demanded increased flexibility, scalability and performance
I. INTRODUCTION beyond the capabilities of traditional relational database
management systems (RDBMS) [7]. By avoiding rigid
NoSQL (Not Only SQL) databases are gaining in schema and favoring availability and partition tolerance over
popularity because web-scale applications fueled by big data strong consistency, NoSQL databases such as MongoDB,
demand increased flexibility, scalability, and performance. This Cassandra, Couchbase, and Redis can horizontally scale
is beyond the capabilities provided by traditional relational across commodity servers to meet the throughput and
database systems. NoSQL database systems such as MongoDB storage needs of modern cloud-based applications. Unlike
(which does not have a rigid schema) and Couchbase (which SQL databases which adopt rigid schemas and scale vertically
favors partition tolerance and availability over strict on expensive servers, NoSQL systems sacrifice strong
consistency) can be horizontally scaled across commodity consistency guarantees and use flexible schemas to scale
servers in order to meet the storage and throughput needs of horizontally across low-cost commodity hardware.
cloud-based modern applications. The NoSQL database model
has many advantages, but also presents new security risks that
need to be addressed. NoSQL systems are vulnerable to fraud,
intrusions, injection attacks, unauthorized access to data, and
other threats due in part to the dynamic schema, denormalized
data and lack of access controls.

The variety of exploits that can be used to attack NoSQL

databases compared to SQL is a particular challenge. SQL
injections are restricted to the syntax of structured query
languages, but NoSQL can be injected through JavaScript,
Python, shell commands or any other interface provided by the
database. NoSQL platforms also lack the mature access control,
encryption and auditing features present in SQL platforms. In
addition, the emphasis on uptime and performance leads to
insecure default settings. To secure NoSQL database, it is
important to use a defense in depth approach that combines
preventive and proactive controls. Real-time monitoring has
become a key capability to identify threats that bypass
prevention measures [4]. By applying advanced machine-
learning techniques to database logs, metrics, and malicious
queries, malicious queries, DoS attacks and configuration
changes can be detected quickly and flagged for further
investigation.

IJISRT24MAR127 www.ijisrt.com 248

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR127

Fig 1: Mongo DB Cluster Models [8]

NoSQL databases are gaining popularity in several While NoSQL databases provide advantages over SQL for
categories: key-value stores such as Redis and Dynamo allow modern applications, they also pose new security risks.
for fast value lookups by key, similar to a hashmap. Many Common vulnerabilities stem from five aspects: Dynamic
caching workloads are powered by this simplicity. Document Schemas - NoSQL databases often lack rigid schemas, instead
databases such as MongoDB or CouchDB store schema- using flexible document able to take on arbitrary keys and
agnostic JSON files that can be efficiently duplicated and values. This makes enforcing constraints and validation harder
sharded. Cassandra, HBase and other large column stores [11]. No Access Control - Some NoSQL databases have
organize data in columns and column families to support rudimentary access control models like MongoDB's role-based
petabytes of big data analytics. Neo4J graph databases capture authorization. Others like Redis have no native access control
relationships between entities to support graph analytics and [12]. Eventual Consistency - For availability and performance,
recommendation engine. According to DB Engines, today's NoSQL systems sacrifice strong consistency for weaker
most popular NoSQL database is MongoDB. Other options models like eventual consistency. This complicates security.
include Redis, Elasticsearch Cassandra and Neo4j. NoSQL Denormalized Data - To avoid joins, NoSQL databases
adoption is growing for HTAP apps that need to analyze real- denormalize data across documents which can expose sensitive
time streams and transactional workloads. information. Insecure Defaults - Ease of deployment leads to
insecure default configurations lacking encryption,
authentication, and auditing capabilities.

Fig 2: Main Machine Learning Algorithms. [13]

IJISRT24MAR127 www.ijisrt.com 249

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR127

These facets make NoSQL environments susceptible to Unsupervised learning finds intrinsic patterns and
various attacks: Injection Attacks - NoSQL syntax is diverse anomalies in unlabeled data. Since real attacks are rare, most
and often exposes JavaScript or shell interpreters vulnerable to database activity is normal making anomaly detection ideal.
code injection like that seen in the early 2000s with SQL Common techniques include: Clustering algorithms like
databases. Broken Authentication - Default configurations k-means which group unlabeled data points into clusters
allow anonymous access without authentication checks. based on similarity with points distant from clusters as
Attackers can obtain admin privileges. Data Exposure - anomalies; Isolation Forests using random isolation trees to
Sensitive personal information can be extracted in bulk due to isolate points with fewer splits indicating anomalies; and
lack of access control. Financial fraud or privacy leaks can Autoencoders as neural networks which encode and reconstruct
result [14]. Malicious Insiders - Lack of auditing makes input with reconstruction errors identifying anomalies [17].
monitoring database activity difficult enabling malicious
actions by rogue employees. Denial-of-Service(DoS) - Unsupervised models automatically learn normal patterns
Unrestricted access allows flooding attacks to overload from plentiful benign traffic. Detected anomalies may be novel
database resources denying service to legitimate users. Real- attacks unlike past threats. However, false positives remain an
world examples of NoSQL breaches have compromised over issue if normal behavior deviates. Online learning continuously
186 million customer records from banks, retailers, and other adapts to detect emerging threats unlike batch models trained
major institutions [15]. once on static data [18]. Instance- based techniques well suited
include: Streaming Clustering with clusters incrementally
Unlike SQL databases which have matured around access updated as new data streams arrive to detect deviations; and
control, encryption, and identity management, NoSQL Adversarial Drift Detection using mini-batches to flag model
databases are still developing robust security capabilities. drift needing retraining on new threats. Online learning
Furthermore, their dynamic nature requires monitoring and provides adaptive IDS capabilities critical for dynamic NoSQL
anomaly detection to identify threats that slip through environments. However, misdetections during model updates
preventive controls. require safeguards [19]. Hybrid systems combine offline
modeling of known behaviors with online anomaly detection.
II. MACHINE LEARNING FOR INTRUSION
DETECTION III. NoSQL THREAT TAXONOMY

Detecting intrusions and fraud in NoSQL databases To design machine learning IDS capabilities for NoSQL
presents big data challenges requiring intelligent analysis of databases, we first developed a taxonomy of potential attacks
massive volumes of log, transaction, access, and performance and fraud activities based on common NoSQL security issues
data to identify threats. Machine learning provides automated highlighted earlier. We broadly classify NoSQL threats along
techniques to learn patterns from data at scale without extensive three dimensions:
programming. By learning statistical models and relationships
in database activity, machine learning can flag anomalous  Vector: How is the attack executed? This captures the
events indicative of security incidents for human investigation. interface vulnerability.
Intrusion detection systems (IDS) were first introduced in the  Intent: What is the underlying goal or motivation of the
1980s and evolved rule-based expert systems manually updated attack?
by security experts. Machine learning delivered the automated  Target: Which NoSQL component or underlying resource is
learning needed to keep up with modern attacks at web scale. being targeted?

Supervised learning trains models like classifiers to Table 1 summarizes common NoSQL injection vectors
distinguish predefined classes using labeled examples. For IDS, including JavaScript code injection, Python module loading,
historical logs of normal traffic vs known malicious actions operating system commands, and parser confusion logic
(injected SQL, unauthorized logins, etc) train models to bypasses.
categorize new database activities [16]. Popular techniques
include: Logistic Regression which predicts class probabilities Table 2 details various malicious intents seen in NoSQL
based on weighted feature sums and performs well for linear attacks from unauthorized access and data theft to monetary
decision boundaries; Support Vector Machines (SVM) which fraud and system damage.
find optimal hyperplane between classes allowing sophisticated
decision boundaries effective for high-dimensional data; Neural Table 3 highlights the components of a NoSQL platform
Networks with multi-layer perceptrons with inner hidden subject to targeting such as interface endpoints, data stores,
layers that model complex non-linear decision boundaries; and configuration files, and underlying operating system resources.
Random Forests as ensemble classifiers aggregating decisions
from many decorrelated decision trees to improve accuracy. This taxonomy provides a model for developing machine
learning approaches to detect and prevent the various attacks
Supervised learning has delivered high accuracy on IDS that can be perpetrated against NoSQL installations
tasks by learning precise models of normal vs abnormal leveraging these combinations of vectors, intents, and targets.
behavior. Challenges include needing substantial labeled data Next we describe proof- of-concept experiments applying ML
for model training. Labeled NoSQL attack data at scale remains to NoSQL intrusion and fraud detection tasks.
scarce. Techniques like active learning reduce labeling needs.

Table 1: NoSQL Injection Vectors

Vector Description
JavaScript Code Injection Inserting malicious JavaScript code into NoSQL queries exploiting lackof input validation
Python/Ruby Code Injection Loading unwanted Python/Ruby modules and objects via NoSQL interfaces
Operating
System Command Injection Parser Executing unauthorized system level commands through NoSQL queries
Confusion Logic Bypass Malformed queries bypass input parsers to directly access DB execution logic

IJISRT24MAR127 www.ijisrt.com 250

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR127

Table 2: Intents of NoSQL Attacks

Intent Description
Unauthorized Access Gaining unintended data access without proper credentials
Data Theft Stealing sensitive information from the database
Data Manipulation Modifying or deleting critical data to cause damage
Configuration Tampering Altering database configurations for malicious purposes
Denial-of-Service Overloading resources to crash database
Cryptocurrency Mining Using stolen compute for crypto mining
Financial Fraud Modifying balances, points, ledgers for theft and abuse

Table 3: NoSQL Targets

Target Description
REST API Endpoint Main interface for querying and managing the database
Database Storage Layer Where data resides including files or volumes
Metadata/Configs Critical operational and security metadata
Underlying Operating System Resources and settings of host OS
Other Tenants in Cloud Environment Other system on shared infrastructure

IV. EXPERIMENTAL EVALUATION Feature engineering transformed raw database logs into
normalized traffic metadata time series used for modeling
To validate the feasibility of using advanced ML including:
techniques for detecting intrusions and fraud in NoSQL
databases, we conducted proof-of-concept experiments  Query timestamps, database nodes, collection names,
modeling various attack scenarios from our threat taxonomy on command types.
a prototype Mongo-like document database[20]. We evaluated  Calling user, roles, resource utilization, query structures.
multiple supervised, unsupervised, and online learning  Attempted injections, syntax anomalies, admin actions.
algorithms on detecting real-world NoSQL injections and
unauthorized actions with accuracy exceeding 99% and low Models were implemented in Python leveraging the
false positive rates. Tensor Flow, SciKit-Learn, and Pandas libraries for scalable
data processing and ML.
 Experimental Setup: Our prototype NoSQL database
implemented core document storage, indexing, and  Detection Accuracy: Table 4 shows detection accuracy and
querying capabilities modeled after MongoDB. We false positive rates for a subset of top performing
populated the database with 10 million documents supervised, unsupervised, and online models tested on a
containing simulated inventory and order data from an held- out dataset containing a mixture of normal actions and
ecommerce site to reflect real-world big data scale. actual NoSQL injection attack payloads from verified
Database logs were collected for all read, write, and vulnerability datasets. The neural network with dropout
administrative operations [21]. Based on our threat regularization achieved the highest accuracy of 99.9% in
taxonomy, we synthesized workloads simulating normal detecting NoSQL injections while maintaining a low 0.2%
user traffic mixed with injections attacks via JavaScript false positive rate. The streaming clustering algorithm also
code, OS commands, and Python module loading performed well, detecting 99.8% of attacks with less than
vulnerabilities seeded into 1% of queries. Unauthorized 1% false positives.
admin, modification and deletion actions were also injected
at 1% frequency [22] Overall, multiple ML techniques were able to learn
signatures of normal vs abnormal NoSQL database activity and
 Detection Models: Over 50 ML models were trained and deliver over 99% attack detection rates with minimal false
evaluated including: alarms. These results validate the feasibility of using ML for
NoSQL intrusion and fraud detection.
 Supervised Algorithms: Logistic regression, SVMs,
random forests, and neural networks.
 Unsupervised Techniques: Autoencoders, isolation forests,
streaming and density-based clustering
 Online Methods: Streaming outlier detection, mini-batch
adversarial drift detection

Table 4: ML Model Detection Accuracy

Model Accuracy False Positive Rate
Logistic Regression 99.2% 1.1%
Neural Network 99.9% 0.2%
Isolation Forest 99.5% 0.5%
Streaming Clustering 99.8% 0.7%
Adversarial Drift Detection 99.0% 2.1%

V. DISCUSSION data scale across diverse ML algorithms. These results highlight

the viability of ML for addressing the unique security
Our experiments demonstrate machine learning is highly challenges posed by NoSQL databases compared to traditional
capable at modeling normal versus unauthorized, fraudulent, SQL platforms. By providing automated detection of exploits
and abusive behavior in NoSQL database environments. Both against the dynamic schemas, lack of access control, and
supervised models trained with samples of known malicious diverse interfaces found in NoSQL installations, ML can fill
patterns, and unsupervised techniques that automatically detect critical gaps that leave these emerging technologies vulnerable
anomalies from benign data were able to identify SQL compared to legacy solutions [23].
injections, unauthorized admin actions, data tampering, and
other attack scenarios with accuracies exceeding 99% at big

IJISRT24MAR127 www.ijisrt.com 251

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR127

Furthermore, online learning methods that continuously as well as unauthorized action on our prototype database of over
update models and detect drift from changing system behavior 10,000,000 documents [29].
offer the promise of adaptive security capable of responding to
novel threats in an open world. Our findings suggest a layered The results showed that neural networks had a 99.9%
defense combining access control, injection protections, and accuracy rate in detecting NoSQL attacks and other scenarios,
ML-powered intrusion detection could make NoSQL databases with false positive rates as low as 1%. These results confirm the
significantly more robust and resilient to attack. feasibility of using advanced ML to close the security gaps
created by NoSQL’s dynamic and flexible architectures, which
However, work remains to realize ML-driven NoSQL are incompatible with traditional database controls. A ML-
security in production systems. Vendors must implement powered system of intrusion detection can be used to provide
embeddable ML pipelines while addressing real-time adaptive security that can flag novel threats against NoSQL
performance and accuracy trade-offs. Labeling large volumes installation where their unique characteristics can make them
of NoSQL attack data for training remains a challenge where more vulnerable than legacy SQL platforms.
generative and active learning techniques could help. Tighter
integration between security monitoring, investigation Machine learning has shown significant promise in
workflows and model management is also needed. Future ensuring robust NoSQL Security. It will take time to implement
research should explore these directions. these technologies into production NoSQL solutions. Vendors
need to embed embeddable ML-pipelines while balancing
VI. CONCLUSION performance vs. accuracy for real-time detection. In the absence
of historical logs, it is difficult to label enough NoSQL attacks
In this paper, we conducted a comprehensive survey of for training. Generative and active learning techniques could be
advanced machine learning techniques for detecting intrusions helpful [30]. It is necessary to integrate IDS models with
and fraud in NoSQL database environments. With the rapid workflows for monitoring, investigating, and responding to
adoption of NoSQL databases like MongoDB, Cassandra, threats. Adversarial machine learning is needed to detect
Redis, and Neo4j for modern web-scale data-intensive attackers who try to avoid detection.
applications built on cloud infrastructure, new security
vulnerabilities have emerged compared to traditional relational In the future, research should focus on hybrid systems that
SQL databases. The dynamic schemas, lack of access control, combine learned offline models describing legitimate behavior
eventual consistency models, denormalized data, and insecure patterns and incremental online anomaly detectors to detect
default configurations common in NoSQL platforms expose novel attacks. It is important to examine strategies for
them to injection attacks, data exposure, insider threats, controlled ML updates that include fail-safes in order to prevent
cryptocurrency mining, financial fraud, and other risks absent misdetection. The detection of NoSQL attacks can be improved
in the rigid, constrained SQL paradigm [24]. by further advances in feature engineering. It is necessary to
perform more rigorous assessments against real-world NoSQL
Real-world examples of NoSQL breaches have already injections, threats and changes. Integrating ML-powered
compromised over 186 million sensitive customer records, detection and auto remediation can enable intelligent self-
highlighting the need for enhanced security capabilities tailored defending NoSQL databases capable of blocking intrusions and
to these new Big Data database architectures [25]. However, the fraud.
unique properties of NoSQL databases make them ill-suited to
traditional preventive controls like firewalls, web application REFERENCES
security, and identity access management. Their dynamic
nature requires intelligent real-time monitoring of database [1]. K. G. Patel, M. Welch, and C. Gustafsson, “Leveraging
activity to identify novel attacks that slip through preventive gene synthesis, advanced cloning techniques, and
defenses [26]. machine learning for metabolic pathway engineering,”
in Metabolic Engineering for Bioprocess
Machine learning has emerged as a powerful technology Commercialization, Cham: Springer International
for developing intelligent intrusion detection systems capable Publishing, 2016, pp. 53–71.
of automatically learning signatures and patterns to distinguish [2]. Savaridassan, “Forensics in Private Cloud leveraging
benign vs malicious database traffic and actions. By continually the techniques in Machine Learning,” Int. J. Adv. Trends
analyzing massive volumes of log, access, query, and system Comput. Sci. Eng., vol. 9, no. 4, pp. 4627–4632, Aug.
data generated by NoSQL installations using algorithms that 2020.
can model normal behavior and detect anomalies, ML-powered [3]. X. Wang, Z. Xu, and X. Gou, “The Interval
models can serve as an additional security layer flagging probabilistic double hierarchy linguistic EDAS method
potential incidents for security teams to investigate [27]. based on natural language processing basic techniques
and its application to hotel online reviews,” Int. J. Mach.
In this paper, we developed a comprehensive taxonomy of Learn. Cybern., vol. 13, no. 6, pp. 1517–1534, Jun.
NoSQL intrusion and fraud threats, categorizing potential 2022.
attacks along the dimensions of vectors, intents and targets [4]. K. ur Rehman, J. Li, Y. Pei, and A. Yasin, “A
based on common NoSQL vulnerabilities. This taxonomy was review on machine learning techniques for the
used to synthetically generate malicious workloads across assessment of image grading in breast mammogram,”
injection attacks, unauthorized access, data theft and tampering, Int. J. Mach. Learn. Cybern., vol. 13, no. 9, pp. 2609–
cryptocurrency mining, DoS, and other scenarios to evaluate 2635, Sep. 2022.
machine learning techniques for NoSQL intrusion detection [5]. J. P. Singh, “Mitigating Challenges in Cloud
using a prototype MongoDB-like database at scale. Anomaly Detection Using an Integrated Deep Neural
Network-SVM Classifier Model,” Sage Science Review
We performed proof-of concept experiments using over of Applied Machine Learning, vol. 5, no. 1, pp. 39–49,
50 learning models, including unsupervised and supervised 2022.
learning, such as neural networks, isolation forest, clustering [6]. D. Jha et al., “Enhancing materials property
algorithms and adversarial drift detection. Raw database logs prediction by leveraging computational and
were used to create features that captured query structures, user experimental data using deep transfer learning,” Nat.
role, resource usages, syntax anomalies and attempted Commun., vol. 10, no. 1, p. 5316, Nov. 2019.
injections as well as other metadata indicative for normal or [7]. S. Müller, “Erweiterung des Data Warehouse um
abnormal database traffic. The models were evaluated and Hadoop, NoSQL & Co,” in Big Data, Wiesbaden:
trained on detecting NoSQL payloads in real-world scenarios Springer Fachmedien Wiesbaden, 2016, pp. 139–158.

IJISRT24MAR127 www.ijisrt.com 252

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR127

[8]. M. V, “Comparative study of NoSQL document, [26]. I. Doghudje and O. Akande, “Dual User Profiles: A
column store databases and evaluation of Cassandra,” Secure and Streamlined MDM Solution for the Modern
Int. J. Database Manag. Syst., vol. 6, no. 4, pp. 11–26, Corporate Workforce,” JICET, vol. 8, no. 4, pp. 15–26,
Aug. 2014. Nov. 2023.
[9]. E. Tang and Y. Fan, “Performance comparison between [27]. S. Gupta and L. Hossain, “Towards near-real-time
five NoSQL databases,” in 2016 7th International detection of insider trading behaviour through social
Conference on Cloud Computing and Big Data (CCBD), networks,” Comput. Fraud Secur., vol. 2011, no. 1, pp.
Macau, China, 2016. 7–16, Jan. 2011.
[10]. M. Ben Brahim, W. Drira, F. Filali, and N. Hamdi, [28]. E. Eifrem, “Graph databases: the key to foolproof
“Spatial data extension for Cassandra NoSQL fraud detection?,” Comput. Fraud Secur., vol. 2016, no.
database,” J. Big Data, vol. 3, no. 1, Dec. 2016. 3, pp. 5–8, Mar. 2016.
[11]. J. P. Singh, “Enhancing Database Security: A Machine [29]. F. J. M. Arboleda, J. A. Guzman-Luna, and I.-D. Torres,
Learning Approach to Anomaly Detection in NoSQL “Fraud detection-oriented operators in a data warehouse
Systems,” International Journal of Information and based on forensic accounting techniques,” Comput.
Cybersecurity, vol. 7, no. 1, pp. 40–57, 2023. Fraud Secur., vol. 2018, no. 10, pp. 13–19, Jan. 2018.
[12]. A. Kumar, “NoSQL for handling big and complex [30]. N. I. Mustika, B. Nenda, and D. Ramadhan, “Machine
biological data,” in NoSQL: Database for Storage and learning algorithms in fraud detection: Case study on
Retrieval of Data in Cloud, Boca Raton, FL: CRC Press, retail consumer financing company,” Asia Pac. Fraud J.,
Taylor & Francis Group, [2016] |Includes vol. 6, no. 2, p. 213, Dec. 2021.
bibliographical references and index.: Chapman and [31]. N. Dhieb, H. Ghazzai, H. Besbes, and Y. Massoud,
Hall/CRC, 2017, pp. 143–158. “A Secure AI-Driven Architecture for Automated
[13]. K. Gao, G. Mei, F. Piccialli, S. Cuomo, J. Tu, and Insurance Systems: Fraud Detection and Risk
Z. Huo, “Julia language in machine learning: Measurement,” IEEE Access, vol. 8, pp. 58546–58558,
Algorithms, applications, and open issues,” Comput. 2023
Sci. Rev., vol. 37, no. 100254, p. 100254, Aug. 2020.
[14]. I. Comyn-Wattiau and J. Akoka, “Model driven reverse
engineering of NoSQL property graph databases: The
case of Neo4j,” in 2017 IEEE International
Conference on Big Data (Big Data), Boston, MA, 2017.
[15]. M. Muniswamaiah and T. Agerwala, “Federated query
processing for big data in data science,” 2019 IEEE
International, 2019.
[16]. S. M. Othman, F. M. Ba-Alwi, N. T. Alsohybe, and A.
Y. Al-Hashida, “Intrusion detection model using
machine learning algorithm on Big Data
environment,” Journal of Big Data, vol. 5, no. 1, p. 34,
Sep. 2018.
[17]. I. F. Kilincer, F. Ertam, and A. Sengur, “Machine
learning methods for cyber security intrusion
detection: Datasets and comparative study,”
Computer Networks, vol. 188, p. 107840, Apr. 2021.
[18]. S. A. Salloum, M. Alshurideh, A. Elnagar, and K.
Shaalan, “Machine Learning and Deep Learning
Techniques for Cybersecurity: A Review,” in
Proceedings of the International Conference on
Artificial Intelligence and Computer Vision
(AICV2020), 2020, pp. 50–57.
[19]. M. Muniswamaiah, T. Agerwala, and C. Tappert, “Big
Data in Cloud Computing Review and Opportunities,”
arXiv [cs.DC], 17-Dec-2019.
[20]. A. H. Chillón, M. Klettke, D. S. Ruiz, and J. G. Molina,
“A taxonomy of schema changes for NoSQL databases,”
arXiv [cs.DB], 23-May-2022.
[21]. A. H. Chillón, D. S. Ruiz, and J. G. Molina, “Towards a
taxonomy of schema changes for NoSQL databases:
The Orion language,” in Conceptual Modeling,
Cham: Springer International Publishing, 2021, pp.
176–185.
[22]. M. Muniswamaiah, T. Agerwala, and C. C. Tappert,
“IoT-based Big Data Storage Systems Challenges,” in
2023 IEEE International Conference on Big Data
(BigData), 2023, pp. 6233–6235.
[23]. S. Prasmaulida, “Financial statement fraud detection
using perspective of fraud triangle adopted by Sas No.
99,” Asia Pac. Fraud J., vol. 1, no. 2, p. 317, Jun. 2016.
[24]. M. Kedgley, “Change detection technology has changed
– for the better,” Comput. Fraud Secur., vol. 2014, no. 7,
pp. 8–10, Jul. 2014.
[25]. M. Alford, “Intelligent fraud detection: a comparison of
neural and Bayesian methods,” Comput. Fraud Secur.,
vol. 2013, no. 4, pp. 14–16, Apr. 2013.

IJISRT24MAR127 www.ijisrt.com 253