By
presenting
an
improved
Intrusion
Detection
System
(IDS)
that
combines
deep
learning
with
support
vector
machines
(SVM),
this
research
increases
network
security.
The
main
goal
is
to
increase
the
accuracy
of
SVM
detection
by
using
a
methodical
feature
selection
and
optimization
technique
tailored
complexity
intrusions.
35
out
42
features
were
chosen
for
RFECV,
algorithmic
in
machine
learning.
To
ensure
preserved
are
those
contribute
most
model's
predictive
capacity
redundant
deleted,
techniques
such
as
RFECV
priority
ranking
ExtraTreesClassifier
take
performance
into
account
during
process.
improve
classifier
performance,
strategies
hyperparameter
tuning
used,
focusing
on
important
data
cutting
down
redundancies.
several
kernel
functions,
including
linear,
polynomial,
RBF,
sigmoid,
compared
study.
Linear
model
combined
was
shown
perform
best.
Our
outperforms
current
IDS
frameworks,
demonstrated
comparative
analysis,
confirming
efficacy
integrating
SVMs
real-time
threat
detection.
KDD
Cup
99
dataset,
which
has
been
widely
used
benchmark
assessing
different
models,
work.
It
offers
consistent,
varied,
large
dataset
so
researchers
may
evaluate
contrast
their
methods.
Researchers
can
experiment
reduction
enhance
because
dataset's
broad
set
features.
FUDMA Journal of Sciences,
Год журнала:
2025,
Номер
9(5), С. 204 - 212
Опубликована: Май 31, 2025
The
rapid
expansion
of
the
Internet
Things
(IoT)
has
vastly
increased
device
connectivity
but
also
expanded
attack
surface.
Resource
constraints
and
heterogeneous
protocols
make
traditional
intrusion
detection
systems
(IDS)
inadequate:
signature-based
methods
miss
novel
threats,
anomaly
detectors
yield
high
false
positive
rates.
We
propose
a
hybrid
model
integrating
CNN,
LSTM,
AdaBoost
for
robust
IoT
detection.
Our
two-stage
pipeline
begins
with
CNN-LSTM
that
automatically
extracts
spatial
temporal
features
from
preprocessed
network
traffic.
CNN
branch
captures
local
patterns,
while
LSTM
models
sequential
traffic
dependencies.
train
on
combined
UNSW-NB15
RT-IoT2022
dataset
205,449
instances
127
initial
features.
Rigorous
preprocessing
(missing-value
imputation,
one-hot
encoding,
Z-score
normalization,
correlation-based
elimination)
reduces
inputs
to
20-feature
subset.
In
second
stage,
we
extract
deep
representations
CNN-LSTM’s
penultimate
layer
input
them
an
classifier
decision-stump
base
learners.
This
ensemble
adaptively
weights
boost
accuracy
controlling
computation.
Experimental
results
show
improved
test
performance:
99.70%
accuracy,
99.90%
precision,
99.78%
recall,
99.84%
F1-score,
2.43%
rate.
These
metrics
outperform
conventional
IDS
(e.g.,
[Churcher
et
al.,
2021:
98.2%
accuracy;
Kumar
98.5%
F1-score]).
model’s
computational
efficiency
during
training
(64
steps/sec)
suggests
potential
scalability,
though
real-world
deployment
validation
remains
future
work.
MATEC Web of Conferences,
Год журнала:
2024,
Номер
392, С. 01089 - 01089
Опубликована: Янв. 1, 2024
The
development
of
new
mobile
communication
and
information
service
technologies
has
opened
up
exciting
possibilities
for
location-based
services.
Users
services
(LBS)
can
access
vital
data
from
their
providers
by
utilizing
location
data.
Maps
navigation,
services,
tourist
social
networking,
many
more
popular
applications
are
available.
A
user's
other
personal
details
must
be
submitted
to
the
in
order
them
work.
For
example,
about
one's
whereabouts
identity.
By
"location
privacy,"
we
mean
idea
that
third
parties
shouldn't
able
track
a
precise
whereabouts.
It
is
important
users'
sensitive
hidden
unauthorized
individuals
when
communicating.
Most
difficult
LBS
concerns
communications
Each
peer
does
duty
reciprocally
collaborative
method,
which
completely
distributed
technique.
most
secure
private
(LBS),
it
employs
cryptographic
methods.
number
people
using
growing
at
rapid
pace
these
days.
At
this
time,
there
isn't
single
method
available
scalability
capabilities.
Building
realistic
computationally
efficient
solution
offers
high
privacy
while
decreasing
processing
overhead
improving
challenging
task.
suggested
cost-effective,
supports
scaling,
highly
resilient
against
security
assaults,
ensures
privacy.
The
dynamic
nature
of
cyber
threats
offers
a
continual
problem
in
the
field
cybersecurity
context
expanding
internet
environment.
This
study
provides
an
in-depth
assessment
literature
on
machine
learning
(ML)
and
deep
(DL)
methodologies
for
network
analysis
intrusion
detection.
review
curates,
assesses,
distils
method-specific
findings
while
considering
temporal
or
thermal
correlations.
It
recognition
importance
data
ML
DL
approaches,
comprehensive
overview
frequently
used
datasets
ML/DL
applications,
as
well
inherent
challenges
adopting
field.
concludes
with
well-informed
recommendations
future
areas
research
this
critical
domine.
International Journal of Innovative Science and Research Technology (IJISRT),
Год журнала:
2024,
Номер
unknown, С. 1839 - 1847
Опубликована: Авг. 5, 2024
A
Convolution
Neural
Network
(CNN)-based
Intrusion
Detection
Model
for
Cyber-attacks
is
of
great
value
in
identifying
and
classifying
attacks
on
any
network.
The
Knowledge
Discovery
Database
Cup
'99
dataset
containing
approximately
4,900,000
single
connection
vectors
was
divided
into
two
phases;
75%
the
total
used
during
learning
process
machine
technique,
while
25%
a
fully
trained
model
to
validate
evaluate
its
performance.
model's
performance
indicated
that
it
can
detect
classify
different
classes
with
an
accuracy
98%
20
epochs
at
0.001
rate
using
learning.
loss
training
validation
7.48%
7.98%,
respectively,
over
epochs,
which
implies
performed
better
dataset.
This
study
demonstrated
convolutional
network-based
classification
shows
high
detection
low
false
negative
rates.
CNN
offers
fidelity
unknown
attacks,
i.e.,
differentiate
between
already-seen
new
zero-day
attacks.
At
end
experiment,
proposed
approach
suitable
modeling
network
IDS
detecting
intrusion
computer
networks
thereby
enabling
secured
environment
proper
functioning
system
