Volume 10 Issue 01-2025
6
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
ABSTRACT
This study investigates the use of artificial intelligence (AI) and machine learning (ML) models to predict, detect, and
mitigate cybersecurity threats, including zero-day attacks, ransomware, and insider threats. Using a comprehensive
dataset of network logs and attack signatures, we evaluated models such as Logistic Regression, Random Forest,
XGBoost, CNN, and LSTMOur results demonstrate that deep learning models, particularly CNN (97.3% AUC-ROC) and
LSTM (96.8% AUC-ROC), significantly outperform traditional methods, excelling in real-time threat detection and
Research Article
LEVERAGING AI AND MACHINE LEARNING FOR PREDICTING,
DETECTING, AND MITIGATING CYBERSECURITY THREATS: A
COMPARATIVE STUDY OF ADVANCED MODELS
Submission Date:
October 25, 2024,
Accepted Date:
December 30, 2024,
Published Date:
January 22, 2025
Crossref Doi:
https://doi.org/10.55640/ijcsis/Volume10Issue01-02
S M Shadul Islam Rishad
Master Of Science in Information Technology, Westcliff University, USA
Farhan Shakil
Master’s
in Cybersecurity Operations, Webster University, Saint Louis, MO, USA
Sanjida Akter Tisha
Master of Science in Information Technology, Washington University of Science and Technology, USA
Sadia Afrin
Department of Computer & Information Science, Gannon University, USA
Md Mehedi Hassan
Master of Science in Information Technology, Washington University of Science and Technology, USA
Mashaeikh Zaman Md. Eftakhar Choudhury
Master of Social Science in Security Studies, Bangladesh University of Professional (BUP), Dhaka
Nabila Rahman
Master’s in information technology management, Webster University, USA
Journal
Website:
https://scientiamreearc
h.org/index.php/ijcsis
Copyright:
Original
content from this work
may be used under the
terms of the creative
commons
attributes
4.0 licence.
Volume 10 Issue 01-2025
7
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
minimizing false positives. This study highlights the practical applicability of AI and ML in enhancing cybersecurity
frameworks, paving the way for more efficient and scalable solutions against evolving threats.
KEYWORDS
Cybersecurity, Artificial Intelligence, Machine Learning, Threat Detection, Zero-Day Attacks, Ransomware, Insider
Threats, AUC-ROC, CNN, LSTM, Deep Learning, Real-Time Monitoring.
INTRODUCTION
Cybersecurity has become a critical challenge in the
modern digital era, where organizations and
individuals face an increasing number of sophisticated
and persistent cyber threats. Attacks such as zero-day
vulnerabilities, ransomware, insider threats, and
advanced persistent threats (APTs) have the potential
to compromise sensitive information, disrupt
operations, and cause severe financial and reputational
damage. Traditional cybersecurity measures, such as
rule-based systems and signature detection, are no
longer sufficient to address the dynamic and evolving
nature of these threats.
Artificial Intelligence (AI) and Machine Learning (ML)
have emerged as transformative technologies in
cybersecurity. These technologies can analyze vast
amounts of data, identify patterns, and predict
potential threats with remarkable accuracy. By
leveraging AI and ML models, organizations can
proactively detect anomalies, mitigate vulnerabilities,
and respond to attacks in real time. This paper explores
how AI and ML models can be utilized to predict,
detect, and mitigate cybersecurity threats, with a focus
on advanced attacks like zero-day vulnerabilities,
ransomware, and insider threats. The study evaluates
the performance of various ML models, including
Logistic Regression, Decision Trees, Random Forest,
CNN, and LSTM, and provides insights into their
effectiveness in real-world applications.
LITERATURE REVIEW
The increasing reliance on digital infrastructure has led
to an alarming rise in cyberattacks. According to a
report by the Ponemon Institute (2023), global
cybercrime is expected to cause damages exceeding
$10.5 trillion annually by 2025. Traditional cybersecurity
measures have relied heavily on rule-based systems,
which, while effective in detecting known threats, fail
to address novel and adaptive attacks such as zero-day
vulnerabilities. As such, there has been a shift towards
employing AI and ML models for proactive and
intelligent threat detection.
AI and ML in Cybersecurity
Recent studies have highlighted the potential of AI and
ML in enhancing cybersecurity frameworks. Nguyen et
al. (2022) demonstrated that ML models, such as
Random Forest and Support Vector Machines, could
detect ransomware attacks with over 90% accuracy.
Similarly, Khan et al. (2021) emphasized the importance
of deep learning techniques, such as CNNs and LSTMs,
in identifying complex patterns within large datasets,
enabling real-time anomaly detection.
Zero-Day Attack Detection
Zero-day attacks are particularly challenging because
they exploit previously unknown vulnerabilities.
Siddiqui et al. (2020) explored the use of anomaly
detection models, such as Autoencoders, to detect
deviations from normal behavior, providing a
Volume 10 Issue 01-2025
8
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
promising solution for identifying zero-day threats.
However, their research highlighted limitations in
scalability and real-time applications, which our study
aims to address.
Ransomware and Insider Threats
Ransomware attacks continue to evolve, employing
sophisticated encryption techniques that make
mitigation increasingly difficult. Research by Zhang et
al. (2023) showed that ensemble models like XGBoost
could achieve high precision in ransomware
classification. On the other hand, insider threats
—
caused by malicious or negligent actions by internal
users
—
are particularly hard to detect. Autoencoders
and Isolation Forests have shown promise in detecting
such threats (Jain et al., 2022), but their effectiveness
diminishes in dynamic environments.
Comparative Studies
Comparative studies have become essential in
identifying the most effective models for cybersecurity
applications. For instance, Gupta and Sharma (2021)
compared traditional algorithms with deep learning
models, finding that CNN and LSTM outperformed
Logistic Regression and Decision Trees in detecting
real-time threats. However, the study lacked a focus on
real-world deployment challenges, which this paper
addresses by evaluating the performance of models in
both controlled and real-life scenarios.
Research Gaps
While existing literature provides valuable insights into
the application of AI and ML in cybersecurity, there is a
lack of comprehensive studies that evaluate multiple
models across diverse threat scenarios. Moreover, the
scalability and practicality of deploying these models in
real-world environments remain underexplored. This
study aims to fill these gaps by conducting an extensive
comparative analysis of traditional, ensemble, and
deep learning models, assessing their performance in
detecting zero-day attacks, ransomware, and insider
threats.
METHODOLOGY
Research Design and Approach
This study adopts a multi-phased, data-driven
experimental research design to evaluate the potential
of artificial intelligence (AI) and machine learning (ML)
models in predicting, detecting, and mitigating
cybersecurity threats such as zero-day attacks,
ransomware, and insider threats. The methodology
emphasizes
the
integration
of
supervised,
unsupervised, and hybrid learning techniques. The
research follows a three-stage process:
1.
Exploratory Analysis: To understand the
underlying patterns and behaviors within the
datasets, exploratory data analysis (EDA) is
conducted to identify correlations, trends, and
anomalies.
2.
Model Development: This phase involves
building and training various AI/ML models to
classify and detect cybersecurity threats.
3.
Evaluation and Mitigation: The final stage
assesses model performance and tests the
efficacy of real-time mitigation strategies, such
as automated responses and preventive
measures.
This structured approach ensures that the study
comprehensively addresses both detection and
mitigation of cyber threats.
Dataset Description
The research uses a combination of publicly available
datasets, proprietary datasets, and synthetically
Volume 10 Issue 01-2025
9
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
generated datasets to capture a wide variety of
cybersecurity threat scenarios. These datasets provide
comprehensive coverage of different attack types,
system behaviors, and user activities. Below in the
table 1 is a detailed overview of the datasets used:
Table 1: Dataset Overview
Dataset
Name
Source
Type
Features
Size
Use Case
Remarks
CICIDS2017
Canadian
Institute for
Cybersecurity
Network
traffic logs
IP
addresses,
ports, protocols,
packet
sizes,
timestamps,
inter-arrival
times
2.8M
records
Zero-day
attack
detection
Provides
labeled
data for normal and
anomalous traffic,
including
DoS,
DDoS, brute force,
etc.
NSL-KDD
UCI Machine
Learning
Repository
Network
traffic and
system logs
Connection
types,
service
types,
flags,
packet lengths,
source
and
destination
addresses
125K
records
General
intrusion
detection
An
improved
version of the KDD
Cup 1999 dataset
addressing
redundancy issues.
CTU-13
University of
the
Czech
Republic
Botnet
traffic
Flow statistics,
DNS
queries,
HTTP requests,
packet headers
13
scenarios
Ransomware
and
botnet
activity
detection
Focuses on botnet
behaviors
and
infected
host
communications.
CERT
Insider
Threat
CERT Division
User
behavior
data
Login/logout
timestamps, file
access,
keystrokes,
emails,
file
downloads,
unusual
patterns
1,000+
users
Insider threat
detection
Simulated dataset
designed to mimic
real-world
insider
threat scenarios.
VirusShare
Dataset
VirusShare
Malware
samples
API
calls,
assembly
instructions, file
hashes,
memory usage
5M+
samples
Ransomware
and malware
detection
Contains
labeled
malware
samples
across
multiple
categories,
including
ransomware
families.
UNSW-
NB15
Australian
Centre
for
Cyber
Security
Network
intrusion
data
Protocols, flow
features,
TCP
flags, anomaly
scores
2.5M
records
Multi-class
threat
classification
Comprehensive
dataset
covering
modern
attacks,
including
fuzzers,
worms,
and
shellcode.
DARPA
2000
Dataset
MIT
Lincoln
Laboratory
Host
and
network
data
Host
logs,
application logs,
3M
records
Detection of
coordinated
attacks
Contains
multi-
phase
attack
scenarios, including
Volume 10 Issue 01-2025
10
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
network
flow
statistics
insider misuse and
external exploits.
Synthetic
Dataset
Generated via
Python scripts
Custom
attack
simulations
File
system
changes,
memory
consumption,
anomalous
system
calls,
and CPU usage
500K
records
Model
validation and
stress testing
Used to test model
robustness
under
highly diverse and
controlled
scenarios.
Phishing
Websites
Dataset
UCI Machine
Learning
Repository
URL
and
website
features
Domain length,
presence of IP
address,
SSL
certificate,
favicon
behavior
11,055
records
Phishing
website
detection
Focused on phishing
attack detection by
analyzing
website
behavior
and
features.
Log4Shell
Dataset
Proprietary
Exploitation
patterns
Java
libraries,
log
messages,
exploit
attempts,
system
responses
100K
records
Zero-day
exploit
detection
Simulated data for
detecting Log4Shell
and other related
zero-day exploits.
Blue Team
Dataset
Custom-
generated
System logs
and alerts
User
authentication
logs,
firewall
alerts, endpoint
protection logs,
security
incident
responses
250K
records
Cybersecurity
incident
management
Designed
to
replicate real-world
blue
team
responses
to
security incidents.
The wide range of datasets ensures robust model
training and testing, covering diverse threat vectors
and real-world scenarios.
Data Preprocessing
We began our data preprocessing phase by addressing
challenges commonly associated with cybersecurity
datasets, such as noise, inconsistencies, missing
values, and class imbalances. Preprocessing was critical
for preparing the data to ensure optimal performance
of our AI and ML models. This phase involved multiple
systematic steps, each tailored to the unique
characteristics of the datasets we used. Below, we
describe these steps in detail.
Data Cleaning
Our first step in preprocessing involved cleaning the
raw datasets. We identified and removed duplicate
records to eliminate redundancy, as duplicated entries
can introduce biases and skew the results. Additionally,
missing values were addressed using a variety of
imputation techniques based on the type and context
of the data. For numerical attributes, we applied mean
and median imputation, while for categorical data, we
employed the most frequent category or K-Nearest
Neighbors (KNN) imputation. Outliers, particularly in
numerical features such as packet size or memory
usage, were carefully detected using statistical
Volume 10 Issue 01-2025
11
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
methods like the Interquartile Range (IQR) and Z-
scores. Outliers were either capped to the 99th
percentile or removed entirely, depending on their
impact on the analysis.
Data Transformation
To prepare the data for our models, we performed
multiple transformations. For categorical variables, we
used encoding techniques. We applied one-hot
encoding to transform nominal variables into binary
features and label encoding for ordinal variables,
ensuring the models could process these variables
efficiently. Temporal features such as timestamps
were converted into time-based metrics like session
durations, daily patterns, and activity frequencies,
which are particularly important for detecting
anomalies and insider threats.
Feature Engineering
Feature engineering played a key role in enhancing the
predictive power of our models. We derived several
new features from existing data to capture underlying
patterns that might not be explicitly represented in the
raw data. For example, in the case of network traffic
datasets, we created features such as protocol
distribution, average packet size per session, and inter-
arrival times. Similarly, for user behavior datasets, we
computed metrics such as the frequency of file access,
login/logout irregularities, and deviations from
baseline behaviors. In malware datasets, we extracted
insights from API call sequences and memory usage
patterns. This systematic feature engineering allowed
us to include domain-specific knowledge and tailor the
input data for each threat category.
Normalization and Scaling
To ensure uniformity across features, we normalized
and scaled the data. For features such as CPU usage,
packet size, and memory consumption, we applied
Min-Max scaling to bring all numerical variables to a
range between 0 and 1. For features requiring
standardization, such as log-transformed values of
network traffic or system resource consumption, we
used Z-score scaling. These techniques ensured that all
features contributed equally to the models and that no
single feature dominated due to scale differences.
Handling Imbalanced Datasets
Class imbalance is a pervasive issue in cybersecurity
datasets, where normal activity often far outweighs
malicious activity. To address this, we employed a
combination of oversampling and under sampling
techniques.
Synthetic
Minority
Oversampling
Technique (SMOTE) was particularly useful for
generating synthetic examples of underrepresented
classes, such as specific types of zero-day attacks or
insider threats. We complemented SMOTE with
random under sampling of the majority class to create
balanced datasets without sacrificing too much data.
Additionally, for multi-class datasets, we used adaptive
resampling techniques to ensure fair representation
across all threat categories.
Data Augmentation
In scenarios where data for certain threat types was
limited, we applied data augmentation techniques. For
instance, in ransomware datasets, we generated new
examples by modifying existing ones, such as altering
file sizes or introducing small variations in file system
changes. Similarly, for user behavior datasets, we
simulated realistic anomalies by inserting synthetic
irregularities into login patterns, file access logs, and
communication metadata. These augmented datasets
allowed us to improve the robustness of our models
against previously unseen variations of threats.
Volume 10 Issue 01-2025
12
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
Dimensionality Reduction
Given the high dimensionality of some datasets, we
applied dimensionality reduction techniques to
optimize
computational
efficiency
without
compromising model performance. We used Principal
Component Analysis (PCA) to reduce the number of
features while preserving the variance in the data.
Additionally, we explored tree-based feature selection
methods, such as Recursive Feature Elimination (RFE)
and feature importance scores derived from Random
Forests, to retain only the most informative features.
These steps ensured that the models could process the
data efficiently and focus on the most relevant
attributes.
Data Splitting
To create training, validation, and test subsets, we
employed stratified sampling techniques. This ensured
that all subsets maintained the original distribution of
classes, particularly for imbalanced datasets. We
allocated 70% of the data to training, 15% to validation,
and 15% to testing. For time-series datasets, such as
those involving sequential logs or network flows, we
applied chronological splitting to maintain the
temporal structure of the data and avoid data leakage
between subsets.
Noise Reduction
Some datasets, particularly those involving user
behavior or network traffic, contained noisy and
irrelevant features that could negatively impact model
performance. We used noise filtering techniques, such
as removing low-variance features and applying
correlation analysis to identify and eliminate highly
correlated variables. Additionally, for datasets
involving text data, such as phishing websites or
malware logs, we applied natural language
preprocessing techniques, such as stop-word removal,
stemming, and lemmatization, to focus on the most
meaningful content.
Encoding Sequence Data
For datasets involving sequences, such as API call
traces in malware detection or command sequences in
insider threat analysis, we applied sequence encoding
techniques. We transformed these sequences into
numerical representations using approaches such as
one-hot encoding, tokenization, and embedding
methods (e.g., Word2Vec and TF-IDF). This allowed us
to capture contextual relationships and patterns
within sequential data effectively.
Balancing Data Consistency Across Datasets
Since we integrated multiple datasets from different
sources, it was critical to ensure consistency across
them. We standardized feature names, aligned time
formats, and ensured uniform labeling conventions.
We also mapped similar features from different
datasets to a common schema, enabling seamless
integration for multi-source analysis.
By meticulously following these preprocessing steps,
we prepared the data to be robust, consistent, and
suitable for training and evaluating our AI and ML
models. These steps were essential for minimizing
biases, improving model accuracy, and ensuring the
validity of our findings.
Model Development
In this phase, we focused on building and fine-tuning
machine learning models to predict, detect, and
mitigate various cybersecurity threats such as zero-day
attacks, ransomware, and insider threats. Our
objective was to develop robust models capable of
learning complex patterns and generalizing effectively
Volume 10 Issue 01-2025
13
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
to unseen data. We systematically approached this
process in several stages, each aimed at optimizing
model performance and aligning it with the specific
requirements of our use cases.
Model Selection
We began by conducting an extensive review of
potential algorithms suited for the detection and
prediction of cybersecurity threats. Given the diverse
nature of our datasets, we selected a combination of
supervised, unsupervised, and ensemble learning
models. For classification tasks, such as detecting
ransomware or insider threats, we focused on
algorithms like Logistic Regression, Decision Trees,
Random Forests, Gradient Boosting Machines (GBM),
XGBoost, and Support Vector Machines (SVM). For
anomaly detection, particularly for identifying zero-day
attacks, we incorporated unsupervised models like
Isolation Forests, Autoencoders, and One-Class SVMs.
Additionally, we explored advanced deep learning
models, such as Convolutional Neural Networks
(CNNs) and Long Short-Term Memory (LSTM)
networks, for sequential data analysis and image-
based malware detection.
Model Training
Once we identified the models, we proceeded to train
them using our preprocessed datasets. We carefully
configured the hyperparameters for each model to
achieve optimal performance. For tree-based
algorithms like Random Forests and XGBoost, we
tuned parameters such as the number of trees,
maximum depth, and learning rate. Similarly, for deep
learning models, we adjusted the number of layers,
neurons, activation functions, and dropout rates. Our
training process leveraged k-fold cross-validation to
ensure the robustness of our models and reduce the
risk of overfitting. In each fold, the training data was
split into k subsets, and the model was trained
iteratively on (k-1) subsets while being validated on the
remaining subset. This approach provided reliable
performance metrics and minimized bias.
Feature Importance Analysis
To enhance the interpretability of our models, we
conducted feature importance analysis. For tree-based
models, we utilized built-in feature importance scores
to identify the most influential variables contributing
to predictions. For deep learning models, particularly
those involving sequential or time-series data, we
employed techniques like SHAP (SHapley Additive
exPlanations) and LIME (Local Interpretable Model-
Agnostic Explanations) to visualize the contribution of
individual features to model decisions. This analysis not
only improved our understanding of the models' inner
workings but also enabled us to refine the feature
engineering process iteratively.
Handling Imbalanced Classes in Training
We recognized the importance of addressing class
imbalances during model development, especially
when training supervised models. For highly
imbalanced datasets, such as those dominated by
benign traffic with few malicious instances, we
employed weighted loss functions to penalize the
model more heavily for misclassifying minority classes.
Additionally, we implemented class-specific sampling
techniques during training to ensure the model learned
adequately from underrepresented threat types.
These strategies were critical in achieving balanced
performance across all classes.
Deep Learning Architecture Design
For deep learning models, we carefully designed
architectures tailored to specific cybersecurity use
cases. For example, in the case of ransomware
Volume 10 Issue 01-2025
14
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
detection using system logs, we used recurrent
architectures like LSTMs and GRUs (Gated Recurrent
Units) to capture temporal dependencies in sequential
data. For image-based malware analysis, we
implemented convolutional layers to extract spatial
features from binary-encoded images. To optimize
model performance, we incorporated regularization
techniques such as dropout, batch normalization, and
early stopping, which helped prevent overfitting and
improved generalization.
Transfer Learning
Given the challenges associated with limited labeled
data in some cases, we utilized transfer learning
techniques. For image-based malware detection, we
fine-tuned pre-trained CNN models like ResNet and
VGG on our datasets. These pre-trained models, which
were originally trained on large-scale image datasets,
allowed
us
to
leverage
existing
feature
representations and significantly reduced training time
while improving performance.
Hyperparameter Tuning
To fine-tune our models, we employed a combination
of grid search and random search techniques. For each
model, we systematically explored a wide range of
hyperparameter values and evaluated their impact on
model performance using validation datasets. For
deep learning models, we also utilized Bayesian
optimization
to
efficiently
navigate
the
hyperparameter search space, focusing on parameters
like learning rates, batch sizes, and optimizer types.
This iterative tuning process ensured that we achieved
the best possible configuration for each model.
Model Ensemble Techniques
To further enhance performance, we explored
ensemble learning methods. By combining predictions
from multiple models, we aimed to capitalize on the
strengths of different algorithms. We implemented
techniques such as bagging, boosting, and stacking.
For instance, in ransomware detection, we combined
the outputs of Random Forests, Gradient Boosting,
and SVMs to create a meta-model that aggregated
predictions and achieved higher accuracy. Similarly, for
anomaly detection, we integrated Autoencoders and
Isolation Forests to improve detection rates for zero-
day attacks.
Model Evaluation
We rigorously evaluated the performance of our
models using a variety of metrics tailored to the
specific requirements of cybersecurity applications.
For binary classification tasks, we focused on metrics
such as precision, recall, F1-score, and area under the
receiver operating characteristic curve (AUC-ROC). For
multi-class classification, we used confusion matrices
and macro-averaged F1-scores to assess performance
across all classes. For anomaly detection tasks,
particularly in zero-day attack identification, we relied
on metrics like precision-recall curves and Matthews
Correlation Coefficient (MCC). Additionally, we
conducted adversarial testing by introducing synthetic
noise and adversarial examples to assess model
robustness against sophisticated attack scenarios.
Iterative Model Refinement
Our model development process was highly iterative.
Based on evaluation results, we continuously refined
the models to address any observed shortcomings. For
instance, if a model showed high false-positive rates in
ransomware detection, we adjusted the decision
thresholds or reweighted the loss function to prioritize
precision. Similarly, for models that struggled with
specific types of insider threats, we revisited feature
engineering and retrained the model using additional
Volume 10 Issue 01-2025
15
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
augmented data. This iterative cycle of training,
evaluation, and refinement allowed us to achieve
consistently high performance across all use cases.
Deployment and Testing
Finally, we prepared the trained models for
deployment in simulated environments. We integrated
the models into a real-time detection pipeline, where
they processed live data streams and generated
predictions in near real-time. Before deployment, we
conducted
extensive
testing
in
controlled
environments, simulating scenarios such as network
intrusions, ransomware outbreaks, and insider threats.
This allowed us to evaluate the models' real-world
performance and fine-tune their parameters for
optimal operation in production settings.
Through this comprehensive model development
process, we successfully created robust AI and ML
models tailored to address the complexities of
cybersecurity threats. These models demonstrated
exceptional performance in both predictive accuracy
and detection speed, making them well-suited for
practical implementation in cybersecurity systems.
RESULTS AND DISCUSSION
In this section, we present the performance results of
the developed models for predicting, detecting, and
mitigating cybersecurity threats. The evaluation
metrics, including accuracy, precision, recall, F1-score,
AUC-ROC, and execution time, were used to
comprehensively assess the effectiveness of each
model. Furthermore, we conducted a comparative
study to determine which model works best in real-
time scenarios and practical applications.
Model Performance Results
After training and testing each model on the
preprocessed
datasets,
we
evaluated
their
performance. The following table summarizes the
results for key metrics across various machine learning
models in the table 2:
Model
Accuracy
(%)
Precision
(%)
Recall
(%)
F1-
Score
(%)
AUC-
ROC (%)
Execution
Time (ms)
Real-Time
Suitability
Logistic
Regression
87.4
85.6
83.2
84.4
88.1
12
Moderate
Decision Tree
82.1
79.8
81.0
80.4
80.2
8
Low
Random Forest
92.3
90.7
91.2
90.9
94.5
25
High
XGBoost
94.1
92.5
92.8
92.6
96.7
30
Very High
Support
Vector
Machine
88.9
86.4
84.7
85.5
89.8
50
Moderate
Isolation Forest
78.5
76.2
74.8
75.5
81.0
20
Moderate
Autoencoder
83.7
80.3
78.5
79.4
85.2
35
Low
CNN
(Deep
Learning)
95.8
94.2
93.5
93.8
97.3
50
Very High
LSTM
(Deep
Learning)
94.7
93.3
92.6
92.9
96.8
55
Very High
Volume 10 Issue 01-2025
16
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
Comparative Study of Models
To determine the best-performing model for real-time
and real-life applications, we evaluated the models not
only based on their performance metrics but also their
ability to process large-scale data efficiently and make
real-time decisions.
1.
Logistic Regression: While Logistic Regression
achieved reasonable accuracy (87.4%) and
precision (85.6%), it struggled with complex,
non-linear
patterns
often
present
in
cybersecurity data. However, its low execution
time (12 ms) made it moderately suitable for
real-time applications where simplicity and
speed are prioritized over accuracy.
2.
Decision Tree: The Decision Tree model
performed the fastest with an execution time
of only 8 ms, but its overall performance
metrics were lower than other models,
especially in recall (81.0%). While suitable for
quick analysis, it is less effective in detecting
complex threats such as zero-day attacks.
3.
Random Forest: Random Forest demonstrated
strong overall performance with an accuracy of
92.3% and an AUC-ROC of 94.5%. Its relatively
low execution time of 25 ms makes it a robust
choice for real-time threat detection, especially
in scenarios where accuracy is critical.
4.
XGBoost: XGBoost outperformed most models
in terms of accuracy (94.1%), precision (92.5%),
and recall (92.8%), making it ideal for
cybersecurity applications. Despite its slightly
higher execution time (30 ms), it remains
highly effective for real-time scenarios due to
its superior handling of imbalanced data and
complex patterns.
5.
Support Vector Machine (SVM): While SVM
provided good accuracy (88.9%) and AUC-ROC
(89.8%), its higher execution time (50 ms)
made it less suitable for real-time applications
compared to Random Forest and XGBoost.
6.
Isolation Forest: As an unsupervised anomaly
detection model, Isolation Forest showed
average performance with an accuracy of
78.5%. Although it is useful for specific use
cases like zero-day attack detection, it lacks the
versatility required for broader cybersecurity
threat detection.
7.
Autoencoder: The Autoencoder achieved
modest results (accuracy of 83.7%) and was
slower than traditional models, with an
execution time of 35 ms. While it is effective for
anomaly detection, it is not ideal for real-time
threat mitigation.
8.
Convolutional Neural Network (CNN): CNN
emerged as one of the top-performing models,
with an accuracy of 95.8% and an AUC-ROC of
97.3%. It is particularly suitable for complex
threat detection, such as image-based malware
classification. However, its higher execution
time (50 ms) limits its applicability in high-
frequency real-time environments.
9.
Long Short-Term Memory (LSTM): LSTM
performed exceptionally well, achieving an
accuracy of 94.7% and an AUC-ROC of 96.8%. Its
strength lies in analyzing sequential and time-
series data, making it highly effective for
detecting insider threats or ransomware.
However, its execution time (55 ms) may be a
limitation in high-speed environments.
Best Model for Real-Time Applications
Volume 10 Issue 01-2025
17
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
Chart 1: Model Evaluation of different model
0
20
40
60
80
100
120
Logistic Regression
Decision Tree
Random Forest
XGBoost
Support Vector Machine
Isolation Forest
Autoencoder
CNN (Deep Learning)
LSTM (Deep Learning)
Model Evaluttion
Real-Time Suitability
Execution Time (ms)
AUC-ROC (%)
F1-Score (%)
Recall (%)
Precision (%)
Accuracy (%)
Volume 10 Issue 01-2025
18
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
Based on our evaluation, XGBoost and Random Forest
were the most effective models for real-time
cybersecurity applications. Both models achieved high
accuracy, precision, and recall while maintaining
relatively low execution times. Random Forest, with its
simplicity and robust performance, is particularly well-
suited for scenarios requiring immediate threat
detection with minimal computational overhead. On
the other hand, XGBoost's advanced capabilities in
handling complex datasets and imbalanced classes
make it ideal for more nuanced cybersecurity
challenges.
Best Model for Real-Life Applications
For broader real-life cybersecurity applications,
including complex threat detection and mitigation,
CNN and LSTM stood out as the best choices. CNN's
ability to process high-dimensional data, such as
malware images, makes it invaluable for malware
analysis. Meanwhile, LSTM's expertise in handling
sequential data makes it highly effective for identifying
insider threats and ransomware patterns. Although
their higher execution times limit their use in real-time
applications, their superior accuracy and reliability
make them indispensable for offline or near-real-time
analysis.
Chart 2 : AUC-ROC curve
This chart represents the AUC-ROC (%) Curve for
various machine learning models used to predict,
detect, and mitigate cybersecurity threats. Here's a
breakdown of the chart and its real-life impact:
Chart Explanation:
1.
X-axis (Models):
Volume 10 Issue 01-2025
19
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
o
It lists the machine learning models used in the
study, such as Logistic Regression, Decision Tree,
Random Forest, XGBoost, etc.
o
Each model represents a specific algorithm with its
strengths
and
weaknesses
in
handling
cybersecurity data.
2.
Y-axis (AUC-ROC in Percentage):
o
The AUC-ROC value indicates the model's
ability to distinguish between malicious and
benign activities in cybersecurity scenarios.
o
A
higher
percentage
means
better
performance in correctly identifying threats
while minimizing false positives and false
negatives.
3.
Key Observations:
o
CNN (Convolutional Neural Networks) achieved
the
highest
AUC-ROC
score
(~97.3%),
demonstrating exceptional performance for
detecting threats in real time. This is likely due to
CNN's ability to analyze complex patterns and
features
in
high-dimensional
cybersecurity
datasets.
o
LSTM (Long Short-Term Memory) also performed
very well (~96.8%), showcasing its strength in
sequential data like log files and time-series data,
which are prevalent in cybersecurity.
o
Traditional models like Decision Tree and Isolation
Forest scored lower, with AUC-ROC percentages
around 80%, indicating limited performance for
complex and dynamic threats like zero-day attacks.
Real-Life Impact:
1.
Enhanced Threat Detection:
o
Models like CNN and LSTM with high AUC-ROC
scores are highly effective in identifying
sophisticated cybersecurity threats, including zero-
day attacks and advanced persistent threats
(APTs).
o
These models can be integrated into real-time
monitoring systems to flag anomalies with higher
accuracy.
2.
Reduced False Positives:
o
A higher AUC-ROC score translates to fewer false
alarms, which reduces the workload on
cybersecurity analysts. This allows teams to focus
on genuine threats, improving operational
efficiency.
3.
Improved Incident Response:
o
Fast and accurate threat detection enables quicker
responses to mitigate potential damage, ensuring
business continuity and safeguarding sensitive
data.
4.
Real-Time Applications:
o
The top-performing models (CNN and LSTM) can
be deployed in Security Information and Event
Management (SIEM) systems, endpoint protection
solutions, and network monitoring tools for real-
time threat analysis.
5.
Model Applicability:
o
While CNNs excel in detecting static and image-like
features, LSTM models are ideal for processing
sequential data like system logs, making them
suitable for dynamic attack patterns.
o
This flexibility allows organizations to select the
appropriate model based on their specific
cybersecurity challenges.
In conclusion, this chart emphasizes the importance of
using advanced machine learning techniques for
Volume 10 Issue 01-2025
20
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
cybersecurity applications. Models with high AUC-ROC
scores provide a reliable and efficient way to protect
against ever-evolving threats, ensuring robust security
frameworks in real-life environments.
CONCLUSION
This study highlights the effectiveness of advanced
machine learning models in detecting, predicting, and
mitigating cybersecurity threats. The comparative
analysis of various algorithms, including Logistic
Regression, Decision Tree, Random Forest, XGBoost,
CNN, and LSTM, demonstrates that deep learning-
based models, particularly CNN and LSTM, significantly
outperform traditional methods. The CNN model
achieved the highest AUC-ROC score (~97.3%),
showcasing its capability to analyze complex patterns
in cybersecurity data, while LSTM (~96.8%) proved
adept at handling sequential data like system logs.
The findings underscore the critical role of machine
learning in modern cybersecurity frameworks. High-
performing models not only enhance threat detection
but also minimize false positives, enabling security
teams to focus on genuine risks. Moreover, these
models can be integrated into real-time systems like
SIEM platforms and endpoint protection solutions,
improving the speed and accuracy of threat mitigation
efforts.
In real-life scenarios, the deployment of CNNs and
LSTMs can significantly strengthen organizations'
defenses against evolving threats, such as zero-day
attacks and advanced persistent threats (APTs). By
leveraging these advanced algorithms, organizations
can ensure robust, proactive security measures,
safeguard
sensitive
information,
and
reduce
operational risks.
As cybersecurity threats continue to evolve, this study
underscores the importance of adopting cutting-edge
machine learning technologies to address these
challenges effectively. Future research should focus on
optimizing these models for resource efficiency,
scalability,
and
applicability
across
diverse
cybersecurity use cases to further advance the field.
ACKNOWLEDGEMENT
All the Authors contributed Equally
REFERENCE
1.
Md Habibur Rahman, Ashim Chandra Das, Md
Shujan Shak, Md Kafil Uddin, Md Imdadul Alam,
Nafis Anjum, Md Nad Vi Al Bony, & Murshida
Alam. (2024). TRANSFORMING CUSTOMER
RETENTION IN FINTECH INDUSTRY THROUGH
PREDICTIVE
ANALYTICS
AND
MACHINE
LEARNING.
The
American
Journal
of
Engineering and Technology, 6(10), 150
–
163.
https://doi.org/10.37547/tajet/Volume06Issue1
0-17
2.
Tauhedur Rahman, Md Kafil Uddin, Biswanath
Bhattacharjee, Md Siam Taluckder, Sanjida
Nowshin Mou, Pinky Akter, Md Shakhaowat
Hossain, Md Rashel Miah, & Md Mohibur
Rahman. (2024). BLOCKCHAIN APPLICATIONS
IN BUSINESS OPERATIONS AND SUPPLY
CHAIN
MANAGEMENT
BY
MACHINE
LEARNING. International Journal of Computer
Science & Information System, 9(11), 17
–
30.
https://doi.org/10.55640/ijcsis/Volume09Issue1
1-03
3.
Md Jamil Ahmmed, Md Mohibur Rahman,
Ashim Chandra Das, Pritom Das, Tamanna
Pervin, Sadia Afrin, Sanjida Akter Tisha, Md
Mehedi Hassan, & Nabila Rahman. (2024).
COMPARATIVE
ANALYSIS
OF
MACHINE
Volume 10 Issue 01-2025
21
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
LEARNING ALGORITHMS FOR BANKING
FRAUD
DETECTION:
A
STUDY
ON
PERFORMANCE, PRECISION, AND REAL-TIME
APPLICATION.
International
Journal
of
Computer Science & Information System, 9(11),
31
–
44.
https://doi.org/10.55640/ijcsis/Volume09Issue1
1-04
4.
Nafis Anjum, Md Nad Vi Al Bony, Murshida
Alam, Mehedi Hasan, Salma Akter, Zannatun
Ferdus, Md Sayem Ul Haque, Radha Das, &
Sadia
Sultana.
(2024).
COMPARATIVE
ANALYSIS OF SENTIMENT ANALYSIS MODELS
ON BANKING INVESTMENT IMPACT BY
MACHINE
LEARNING
ALGORITHM.
International Journal of Computer Science &
Information
System,
9(11),
5
–
16.
https://doi.org/10.55640/ijcsis/Volume09Issue1
1-02
5.
Das, A. C., Mozumder, M. S. A., Hasan, M. A.,
Bhuiyan, M., Islam, M. R., Hossain, M. N., ... &
Alam, M. I. (2024). MACHINE LEARNING
APPROACHES FOR DEMAND FORECASTING:
THE IMPACT OF CUSTOMER SATISFACTION ON
PREDICTION ACCURACY. The American Journal
of Engineering and Technology, 6(10), 42-53.
6.
Akter, S., Mahmud, F., Rahman, T., Ahmmed,
M. J., Uddin, M. K., Alam, M. I., ... & Jui, A. H.
(2024). A COMPREHENSIVE STUDY OF
MACHINE LEARNING APPROACHES FOR
CUSTOMER
SENTIMENT
ANALYSIS
IN
BANKING SECTOR. The American Journal of
Engineering and Technology, 6(10), 100-111.
7.
Md Risalat Hossain Ontor, Asif Iqbal, Emon
Ahmed,
Tanvirahmedshuvo,
&
Ashequr
Rahman.
(2024).
LEVERAGING
DIGITAL
TRANSFORMATION AND SOCIAL MEDIA
ANALYTICS FOR OPTIMIZING US FASHION
BRANDS’
PERFORMANCE:
A
MACHINE
LEARNING APPROACH. International Journal
of Computer Science & Information System,
9(11),
45
–
56.
https://doi.org/10.55640/ijcsis/Volume09Issue1
1-05
8.
Rahman, A., Iqbal, A., Ahmed, E., & Ontor, M. R.
H. (2024). PRIVACY-PRESERVING MACHINE
LEARNING: TECHNIQUES, CHALLENGES, AND
FUTURE DIRECTIONS IN SAFEGUARDING
PERSONAL
DATA
MANAGEMENT.
International
journal
of
business
and
management sciences, 4(12), 18-32.
9.
Md Jamil Ahmmed, Md Mohibur Rahman,
Ashim Chandra Das, Pritom Das, Tamanna
Pervin, Sadia Afrin, Sanjida Akter Tisha, Md
Mehedi Hassan, & Nabila Rahman. (2024).
COMPARATIVE
ANALYSIS
OF
MACHINE
LEARNING ALGORITHMS FOR BANKING
FRAUD
DETECTION:
A
STUDY
ON
PERFORMANCE, PRECISION, AND REAL-TIME
APPLICATION.
International
Journal
of
Computer Science & Information System, 9(11),
31
–
44.
https://doi.org/10.55640/ijcsis/Volume09Issue1
1-04
10.
Arif, M., Ahmed, M. P., Al Mamun, A., Uddin, M.
K., Mahmud, F., Rahman, T., ... & Helal, M.
(2024). DYNAMIC PRICING IN FINANCIAL
TECHNOLOGY:
EVALUATING
MACHINE
LEARNING
SOLUTIONS
FOR
MARKET
ADAPTABILITY. International Interdisciplinary
Business Economics Advancement Journal,
5(10), 13-27.
11.
Uddin, M. K., Akter, S., Das, P., Anjum, N.,
Akter, S., Alam, M., ... & Pervin, T. (2024).
MACHINE
LEARNING-BASED
EARLY
DETECTION
OF
KIDNEY
DISEASE:
A
COMPARATIVE
STUDY
OF
PREDICTION
MODELS AND PERFORMANCE EVALUATION.
Volume 10 Issue 01-2025
22
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
International Journal of Medical Science and
Public Health Research, 5(12),58-75.
12.
Ponemon Institute. (2023). Cost of a Data
Breach Report 2023. IBM Security. Retrieved
from
https://www.ibm.com/security/data-
breach.
13.
Nguyen, T., Le, Q., & Vu, H. (2022). Machine
learning for ransomware detection: A
comparative analysis of algorithms. Journal of
Cybersecurity Research, 15(3), 105
–
120.
14.
Khan, M. U., Zafar, A., & Ahmed, I. (2021). Deep
learning in cybersecurity: Applications and
challenges. IEEE Transactions on Information
Forensics and Security, 16(8), 1875
–
1890.
https://doi.org/10.1109/TIFS.2021.3051148.
15.
Siddiqui, F., Alam, S., & Hassan, M. (2020).
Detecting zero-day attacks using anomaly-
based machine learning models. Journal of
Network Security, 12(4), 312
–
330.
16.
Zhang, Y., Li, X., & Wang, J. (2023).
Classification of ransomware using ensemble
machine learning models. Computers &
Security,
122,
102759.
https://doi.org/10.1016/j.cose.2023.102759.
17.
Jain, P., Gupta, R., & Kaur, H. (2022). Detection
of insider threats using anomaly-based
machine learning techniques. Cybersecurity
and
Privacy,
1(2),
210
–
225.
https://doi.org/10.3390/cybersecurity1020015.
18.
Gupta, R., & Sharma, A. (2021). A comparative
study of machine learning models for real-time
cybersecurity threat detection. International
Journal of Computer Applications, 183(7), 17
–
25. https://doi.org/10.5120/ijca202118307.
19.
Zhang, J., Wu, T., & Lin, H. (2020). Application
of convolutional neural networks in detecting
advanced persistent threats. IEEE Access, 8,
49332
–
49345.
https://doi.org/10.1109/ACCESS.2020.2981249.
20.
Jain, A., Verma, K., & Sinha, S. (2021). Real-time
applications of LSTM for analyzing time-series
data in cybersecurity. Expert Systems with
Applications,
165,
113898.
https://doi.org/10.1016/j.eswa.2020.113898
21.
Shak, M. S., Uddin, A., Rahman, M. H., Anjum,
N., Al Bony, M. N. V., Alam, M., ... & Pervin, T.
(2024). INNOVATIVE MACHINE LEARNING
APPROACHES
TO
FOSTER
FINANCIAL
INCLUSION IN MICROFINANCE. International
Interdisciplinary
Business
Economics
Advancement Journal, 5(11), 6-20.
22.
Naznin, R., Sarkar, M. A. I., Asaduzzaman, M.,
Akter, S., Mou, S. N., Miah, M. R., ... & Sajal, A.
(2024).
ENHANCING
SMALL
BUSINESS
MANAGEMENT
THROUGH
MACHINE
LEARNING: A COMPARATIVE STUDY OF
PREDICTIVE
MODELS
FOR
CUSTOMER
RETENTION, FINANCIAL FORECASTING, AND
INVENTORY OPTIMIZATION. International
Interdisciplinary
Business
Economics
Advancement Journal, 5(11), 21-32.
23.
Bhattacharjee, B., Mou, S. N., Hossain, M. S.,
Rahman, M. K., Hassan, M. M., Rahman, N., ...
& Haque, M. S. U. (2024). MACHINE LEARNING
FOR COST ESTIMATION AND FORECASTING IN
BANKING: A COMPARATIVE ANALYSIS OF
ALGORITHMS.
Frontline
Marketing,
Management and Economics Journal, 4(12), 66-
83.
24.
Rahman, A., Iqbal, A., Ahmed, E., & Ontor, M. R.
H. (2024). PRIVACY-PRESERVING MACHINE
LEARNING: TECHNIQUES, CHALLENGES, AND
FUTURE DIRECTIONS IN SAFEGUARDING
PERSONAL DATA MANAGEMENT. Frontline
Marketing, Management and Economics
Journal, 4(12), 84-106.
25.
Al Mamun, A., Hossain, M. S., Rishad, S. S. I.,
Rahman, M. M., Shakil, F., Choudhury, M. Z. M.
Volume 10 Issue 01-2025
23
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
E., ... & Sultana, S. (2024). MACHINE LEARNING
FOR
STOCK
MARKET
SECURITY
MEASUREMENT: A COMPARATIVE ANALYSIS
OF SUPERVISED, UNSUPERVISED, AND DEEP
LEARNING MODELS. The American Journal of
Engineering and Technology, 6(11), 63-76.
26.
Das, A. C., Rishad, S. S. I., Akter, P., Tisha, S. A.,
Afrin, S., Shakil, F., ... & Rahman, M. M. (2024).
ENHANCING BLOCKCHAIN SECURITY WITH
MACHINE LEARNING: A COMPREHENSIVE
STUDY OF ALGORITHMS AND APPLICATIONS.
The American Journal of Engineering and
Technology, 6(12), 150-162.
27.
Rahman, M. M., Akhi, S. S., Hossain, S., Ayub,
M. I., Siddique, M. T., Nath, A., ... & Hassan, M.
M. (2024). EVALUATING MACHINE LEARNING
MODELS
FOR
OPTIMAL
CUSTOMER
SEGMENTATION
IN
BANKING:
A
COMPARATIVE STUDY. The American Journal
of Engineering and Technology, 6(12), 68-83.
28.
Das, P., Pervin, T., Bhattacharjee, B., Karim, M.
R., Sultana, N., Khan, M. S., ... & Kamruzzaman,
F. N. U. (2024). OPTIMIZING REAL-TIME
DYNAMIC PRICING STRATEGIES IN RETAIL AND
E-COMMERCE USING MACHINE LEARNING
MODELS. The American Journal of Engineering
and Technology, 6(12), 163-177.
29.
Hossain, M. N., Hossain, S., Nath, A., Nath, P. C.,
Ayub, M. I., Hassan, M. M., ... & Rasel, M.
(2024).
ENHANCED
BANKING
FRAUD
DETECTION: A COMPARATIVE ANALYSIS OF
SUPERVISED
MACHINE
LEARNING
ALGORITHMS. American Research Index
Library, 23-35.
30.
Shak, M. S., Mozumder, M. S. A., Hasan, M. A.,
Das, A. C., Miah, M. R., Akter, S., & Hossain, M.
N. (2024). OPTIMIZING RETAIL DEMAND
FORECASTING:
A
PERFORMANCE
EVALUATION
OF
MACHINE
LEARNING
MODELS INCLUDING LSTM AND GRADIENT
BOOSTING.
The
American
Journal
of
Engineering and Technology, 6(09), 67-80.
31.
Das, A. C., Mozumder, M. S. A., Hasan, M. A.,
Bhuiyan, M., Islam, M. R., Hossain, M. N., ... &
Alam, M. I. (2024). MACHINE LEARNING
APPROACHES FOR DEMAND FORECASTING:
THE IMPACT OF CUSTOMER SATISFACTION ON
PREDICTION ACCURACY. The American Journal
of Engineering and Technology, 6(10), 42-53.
32.
Hossain, M. N., Anjum, N., Alam, M., Rahman,
M. H., Taluckder, M. S., Al Bony, M. N. V., ... &
Jui, A. H. (2024). PERFORMANCE OF MACHINE
LEARNING ALGORITHMS FOR LUNG CANCER
PREDICTION:
A
COMPARATIVE
STUDY.
International Journal of Medical Science and
Public Health Research, 5(11), 41-55.
33.
Miah, J., Khan, R. H., Ahmed, S., & Mahmud, M.
I. (2023, June). A comparative study of
detecting covid 19 by using chest X-ray images
–
A deep learning approach. In 2023 IEEE World
AI IoT Congress (AIIoT) (pp. 0311-0316). IEEE.
34.
Khan, R. H., Miah, J., Nipun, S. A. A., & Islam, M.
(2023, March). A Comparative Study of
Machine Learning classifiers to analyze the
Precision of Myocardial Infarction prediction.
In 2023 IEEE 13th Annual Computing and
Communication Workshop and Conference
(CCWC) (pp. 0949-0954). IEEE.
35.
Kayyum, S., Miah, J., Shadaab, A., Islam, M. M.,
Islam, M., Nipun, S. A. A., ... & Al Faisal, F. (2020,
January). Data analysis on myocardial
infarction with the help of machine learning
algorithms considering distinctive or non-
distinctive features. In 2020 International
Conference on Computer Communication and
Informatics (ICCCI) (pp. 1-7). IEEE.
36.
Islam, M. M., Nipun, S. A. A., Islam, M., Rahat,
M. A. R., Miah, J., Kayyum, S., ... & Al Faisal, F.
Volume 10 Issue 01-2025
24
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
(2020). An empirical study to predict
myocardial infarction using k-means and
hierarchical clustering. In Machine Learning,
Image Processing, Network Security and Data
Sciences: Second International Conference,
MIND 2020, Silchar, India, July 30-31, 2020,
Proceedings, Part II 2 (pp. 120-130). Springer
Singapore.
37.
Miah, J., Ca, D. M., Sayed, M. A., Lipu, E. R.,
Mahmud, F., & Arafat, S. Y. (2023, November).
Improving Cardiovascular Disease Prediction
Through Comparative Analysis of Machine
Learning Models: A Case Study on Myocardial
Infarction.
In
2023
15th
International
Conference on Innovations in Information
Technology (IIT) (pp. 49-54). IEEE.
38.
Khan, R. H., Miah, J., Rahat, M. A. R., Ahmed, A.
H., Shahriyar, M. A., & Lipu, E. R. (2023,
September). A Comparative Analysis of
Machine Learning Approaches for Chronic
Kidney Disease Detection. In 2023 8th
International
Conference
on
Electrical,
Electronics and Information Engineering
(ICEEIE) (pp. 1-6). IEEE.
39.
Miah, J., Cao, D. M., Sayed, M. A., Taluckder, M.
S., Haque, M. S., & Mahmud, F. (2023).
Advancing Brain Tumor Detection: A Thorough
Investigation of CNNs, Clustering, and SoftMax
Classification in the Analysis of MRI Images.
arXiv preprint arXiv:2310.17720.
40.
Rahman, M. M., Islam, A. M., Miah, J., Ahmad,
S., & Mamun, M. (2023, June). sleepWell: Stress
Level Prediction Through Sleep Data. Are You
Stressed?. In 2023 IEEE World AI IoT Congress
(AIIoT) (pp. 0229-0235). IEEE.
41.
Rahman, M. M., Islam, A. M., Miah, J., Ahmad,
S., & Hasan, M. M. (2023, June). Empirical
Analysis with Component Decomposition
Methods for Cervical Cancer Risk Assessment.
In 2023 IEEE World AI IoT Congress (AIIoT) (pp.
0513-0519). IEEE.
42.
Khan, R. H., Miah, J., Nipun, S. A. A., Islam, M.,
Amin, M. S., & Taluckder, M. S. (2023,
September). Enhancing Lung Cancer Diagnosis
with
Machine
Learning
Methods
and
Systematic Review Synthesis. In 2023 8th
International
Conference
on
Electrical,
Electronics and Information Engineering
(ICEEIE) (pp. 1-5). IEEE.
43.
Miah, J. (2024). HOW FAMILY DNA CAN CAUSE
LUNG CANCER USING MACHINE LEARNING.
International Journal of Medical Science and
Public Health Research, 5(12), 8-14.
44.
Miah, J., Khan, R. H., Linkon, A. A., Bhuiyan, M.
S., Jewel, R. M., Ayon, E. H., ... & Tanvir Islam,
M. (2024). Developing a Deep Learning
Methodology to Anticipate the Onset of
Diabetic Retinopathy at an Early Stage. In
Innovative and Intelligent Digital Technologies;
Towards an Increased Efficiency: Volume 1 (pp.
77-91). Cham: Springer Nature Switzerland.
45.
Hasan, M., Pathan, M. K. M., & Kabir, M. F.
(2024). Functionalized Mesoporous Silica
Nanoparticles as Potential Drug Delivery
Vehicle against Colorectal Cancer. Journal of
Medical and Health Studies, 5(3), 56-62.
46.
Hasan, M. (2023). SURFACE-ENGINEERED
MINERAL PARTICLES FOR GATED DRUG
DELIVERY, GENE TRANSFER AND SUNSCREEN
FORMULATIONS.
47.
Hasan, M., Kabir, M. F., & Pathan, M. K. M.
(2024). PEGylation of Mesoporous Silica
Nanoparticles for Drug Delivery Applications.
Journal of Chemistry Studies, 3(2), 01-06.
48.
Hasan, M., Evett, C. G., & Burton, J. (2024).
Advances in Nanoparticle-Based Targeted Drug
Delivery Systems for Colorectal Cancer
Volume 10 Issue 01-2025
25
INTERNATIONAL JOURNAL OF COMPUTER SCIENCE & INFORMATION
SYSTEM (eISSN: 2536-7919 pISSN: 2536-7900)
VOLUME
10
ISSUE
01
Pages:
6-25
Therapy:
A
Review.
arXiv
preprint
arXiv:2409.05222.
49.
Hasan, M., & Mahama, M. T. (2024).
Uncovering the complex mechanisms behind
nanomaterials-based
plasmon-driven
photocatalysis through the utilization of
Surface-Enhanced Raman Spectroscopies.
arXiv preprint arXiv:2408.13927.
