IETE Journal of Research,
Год журнала:
2024,
Номер
unknown, С. 1 - 12
Опубликована: Ноя. 24, 2024
Cloud
computing
(CC)
is
one
of
the
most
promising
technologies
for
effectively
storing
data
and
offering
internet
services.
There
are
several
benefits
to
using
this
quickly
evolving
technology
instead
more
conventional
defenses
shield
computer-based
systems
from
cyberattacks.
Nevertheless,
there
a
number
drawbacks
current
approaches,
including
limited
precision,
which
may
affect
system
performance,
scalability,
security,
efficiency.
To
overcome
these
issues,
novel
spectral
recurrent
neural
network
(RNN)-based
intrusion
detection
in
cloud
environment
has
been
proposed
improve
security
CC.
Initially,
preprocessing
uses
IoT-23
dataset
values
reduce
null
or
inappropriate
feature
values.
Then
fuzzified
entity
used
selection
analyze
features
based
on
its
threshold
Using
support
index
behavioral
rate,
select
relational
their
maximum
range.
Soft-max
deep
RNN
identifying
non-intrusion.
The
intruded
classified
recursive
multi-perception
classifier
categorized
risk
level.
model
superior
accuracy,
recall,
specificity,
F
measure,
according
experimental
data.
suggested
tactics
provide
outcomes
with
an
accuracy
rate
99.14%,
notable
enhancement
over
previous
studies
substantiates
effectiveness
methodology.
Scientific Reports,
Год журнала:
2025,
Номер
15(1)
Опубликована: Янв. 14, 2025
Smart
devices
are
enabled
via
the
Internet
of
Things
(IoT)
and
connected
in
an
uninterrupted
world.
These
pose
a
challenge
to
cybersecurity
systems
due
attacks
network
communications.
Such
have
continued
threaten
operation
end-users.
Therefore,
Intrusion
Detection
Systems
(IDS)
remain
one
most
used
tools
for
maintaining
such
flaws
against
cyber-attacks.
The
dynamic
multi-dimensional
threat
landscape
IoT
increases
Traditional
IDS.
focus
this
paper
aims
find
key
features
developing
IDS
that
is
reliable
but
also
efficient
terms
computation.
Enhanced
Grey
Wolf
Optimization
(EGWO)
Feature
Selection
(FS)
implemented.
function
EGWO
remove
unnecessary
from
datasets
intrusion
detection.
To
test
new
FS
technique
decide
on
optimal
set
based
accuracy
achieved
feature
taking
filters,
recent
approach
relies
NF-ToN-IoT
dataset.
selected
evaluated
by
using
Random
Forest
(RF)
algorithm
combine
multiple
decision
trees
create
accurate
result.
experimental
outcomes
procedures
demonstrate
capacity
recommended
classification
methods
determine
Analysis
results
presents
performs
more
effectively
than
other
techniques
with
optimized
(i.e.,
23
out
43
features),
high
99.93%
improved
convergence.
Algorithms,
Год журнала:
2025,
Номер
18(2), С. 69 - 69
Опубликована: Янв. 28, 2025
As
security
threats
become
more
complex,
the
need
for
effective
intrusion
detection
systems
(IDSs)
has
grown.
Traditional
machine
learning
methods
are
limited
by
extensive
feature
engineering
and
data
preprocessing.
To
overcome
this,
we
propose
two
enhanced
hybrid
deep
models,
an
autoencoder–convolutional
neural
network
(Autoencoder–CNN)
a
transformer–deep
(Transformer–DNN).
The
Autoencoder
reshapes
traffic
data,
addressing
class
imbalance,
CNN
performs
precise
classification.
transformer
component
extracts
contextual
features,
which
DNN
uses
accurate
Our
approach
utilizes
adaptive
synthetic
sampling–synthetic
minority
oversampling
technique
(ADASYN-SMOTE)
binary
classification
SMOTE
multi-class
classification,
along
with
edited
nearest
neighbors
(ENN)
further
imbalance
handling.
models
were
designed
to
minimize
false
positives
negatives,
improve
real-time
detection,
identify
zero-day
attacks.
Evaluations
based
on
CICIDS2017
dataset
showed
99.90%
accuracy
Autoencoder–CNN
99.92%
Transformer–DNN
in
99.95%
99.96%
respectively.
On
NF-BoT-IoT-v2
dataset,
achieved
99.98%
97.95%
while
reached
97.90%,
These
results
demonstrate
superior
performance
of
proposed
compared
traditional
handling
diverse
Wiley Interdisciplinary Reviews Data Mining and Knowledge Discovery,
Год журнала:
2025,
Номер
15(2)
Опубликована: Март 28, 2025
ABSTRACT
As
the
Internet
of
Things
(IoT)
continues
expanding
its
footprint
across
various
sectors,
robust
security
systems
to
mitigate
associated
risks
are
more
critical
than
ever.
Intrusion
Detection
Systems
(IDS)
fundamental
in
safeguarding
IoT
infrastructures
against
malicious
activities.
This
systematic
review
aims
guide
future
research
by
addressing
six
pivotal
questions
that
underscore
development
advanced
IDS
tailored
for
environments.
Specifically,
concentrates
on
applying
machine
learning
(ML)
and
deep
(DL)
technologies
enhance
capabilities.
It
explores
feature
selection
methodologies
aimed
at
developing
lightweight
solutions
both
effective
efficient
scenarios.
Additionally,
assesses
different
datasets
balancing
techniques,
which
crucial
training
models
perform
accurately
reliably.
Through
a
comprehensive
analysis
existing
literature,
this
highlights
significant
trends,
identifies
current
gaps,
suggests
studies
optimize
frameworks
ever‐evolving
landscape.
Future Internet,
Год журнала:
2024,
Номер
16(7), С. 253 - 253
Опубликована: Июль 18, 2024
The
proliferation
of
IoT
services
has
spurred
a
surge
in
network
attacks,
heightening
cybersecurity
concerns.
Essential
to
defense,
intrusion
detection
and
prevention
systems
(IDPSs)
identify
malicious
activities,
including
denial
service
(DoS),
distributed
(DDoS),
botnet,
brute
force,
infiltration,
Heartbleed.
This
study
focuses
on
leveraging
unsupervised
learning
for
training
models
counter
these
threats
effectively.
proposed
method
utilizes
basic
autoencoders
(bAEs)
dimensionality
reduction
encompasses
three-stage
model:
one-class
support
vector
machine
(OCSVM)
deep
autoencoder
(dAE)
attack
detection,
complemented
by
density-based
spatial
clustering
applications
with
noise
(DBSCAN)
clustering.
Accurately
delineated
clusters
aid
mapping
tactics.
MITRE
ATT&CK
framework
establishes
“Cyber
Threat
Repository”,
cataloging
attacks
tactics,
enabling
immediate
response
based
priority.
Leveraging
preprocessed
unlabeled
normal
traffic
data,
this
approach
enables
the
identification
novel
while
mitigating
impact
imbalanced
data
model
performance.
reconstruction
error,
OCSVM
employs
kernel
function
establish
hyperplane
anomaly
DBSCAN
clusters,
manage
noise,
accommodate
diverse
shapes,
automatically
determining
cluster
count,
ensuring
scalability,
minimizing
false
positives
negatives.
Evaluated
standard
datasets
such
as
CIC-IDS2017
CSECIC-IDS2018,
outperforms
existing
state
art
methods.
Our
achieves
accuracies
exceeding
98%
two
datasets,
thus
confirming
its
efficacy
effectiveness
application
efficient
systems.
Future Internet,
Год журнала:
2024,
Номер
16(12), С. 481 - 481
Опубликована: Дек. 23, 2024
Network
and
cloud
environments
must
be
fortified
against
a
dynamic
array
of
threats,
intrusion
detection
systems
(IDSs)
are
critical
tools
for
identifying
thwarting
hostile
activities.
IDSs,
classified
as
anomaly-based
or
signature-based,
have
increasingly
incorporated
deep
learning
models
into
their
framework.
Recently,
significant
advancements
been
made
in
particularly
those
using
machine
learning,
where
attack
accuracy
has
notably
high.
Our
proposed
method
demonstrates
that
can
achieve
unprecedented
success
both
known
unknown
threats
within
environments.
However,
existing
benchmark
datasets
typically
contain
more
normal
traffic
samples
than
to
reflect
real-world
network
traffic.
This
imbalance
the
training
data
makes
it
challenging
IDSs
accurately
detect
specific
types
attacks.
Thus,
our
challenges
arise
from
two
key
factors,
unbalanced
emergence
new,
unidentified
threats.
To
address
these
issues,
we
present
hybrid
transformer-convolutional
neural
(Transformer-CNN)
model,
which
leverages
resampling
techniques
such
adaptive
synthetic
(ADASYN),
minority
oversampling
technique
(SMOTE),
edited
nearest
neighbors
(ENN),
class
weights
overcome
imbalance.
The
transformer
component
model
is
employed
contextual
feature
extraction,
enabling
system
analyze
relationships
patterns
effectively.
In
contrast,
CNN
responsible
final
classification,
processing
extracted
features
identify
types.
Transformer-CNN
focuses
on
three
primary
objectives
enhance
performance:
(1)
reducing
false
positives
negatives,
(2)
real-time
high-speed
networks,
(3)
detecting
zero-day
We
evaluate
Transformer-CNN,
NF-UNSW-NB15-v2
CICIDS2017
datasets,
assess
its
performance
with
metrics
accuracy,
precision,
recall,
F1-score.
results
demonstrate
achieves
an
impressive
99.71%
binary
classification
99.02%
multi-class
dataset,
while
reaches
99.93%
99.13%
significantly
outperforming
models.
proves
enhanced
capability
IDS
defending
intrusions,
including
Sensors,
Год журнала:
2025,
Номер
25(3), С. 847 - 847
Опубликована: Янв. 30, 2025
The
rise
in
intrusions
on
network
and
IoT
systems
has
led
to
the
development
of
artificial
intelligence
(AI)
methodologies
intrusion
detection
(IDSs).
However,
traditional
AI
or
machine
learning
(ML)
methods
can
compromise
accuracy
due
vast,
diverse,
dynamic
nature
data
generated.
Moreover,
many
these
lack
transparency,
making
it
challenging
for
security
professionals
make
predictions.
To
address
challenges,
this
paper
presents
a
novel
IDS
architecture
that
uses
deep
(DL)-based
methodology
along
with
eXplainable
(XAI)
techniques
create
explainable
models
systems,
empowering
analysts
use
effectively.
DL
are
needed
train
enormous
amounts
produce
promising
results.
Three
different
models,
i.e.,
customized
1-D
convolutional
neural
networks
(1-D
CNNs),
(DNNs),
pre-trained
model
TabNet,
proposed.
experiments
performed
seven
datasets
TON_IOT.
CNN
dataset
achieves
an
impressive
99.24%.
Meanwhile,
six
datasets,
most
DNN
achieve
100%
accuracy,
further
validating
effectiveness
proposed
models.
In
all
least-performing
is
TabNet.
Implementing
method
real
time
requires
explanation
predictions
Thus,
XAI
implemented
understand
essential
features
responsible
predicting
particular
class.