PLoS ONE,
Journal Year:
2024,
Volume and Issue:
19(11), P. e0310840 - e0310840
Published: Nov. 1, 2024
Over
the
last
25
years,
a
considerable
proliferation
of
software
metrics
and
plethora
tools
have
emerged
to
extract
them.
While
this
is
indeed
positive
concerning
previous
situations
limited
data,
it
still
leads
significant
problem
arising
both
from
theoretical
practical
standpoint.
From
perspective,
several
are
likely
result
in
collinearity,
overfitting,
etc.
such
set
difficult
manage
companies,
especially
small
ones,
may
feel
overwhelmed
unable
select
viable
subset
Still,
so
far
has
not
been
fully
understood
what
suitable
properly
projects
products.
In
paper,
we
attempt
address
issue.
We
focus
on
case
programs
written
Java
consider
classes
methods.
use
Sammon
error
as
measure
similarity
metrics.
Utilizing
Particle
Swarm
Optimization
Genetic
Algorithm,
adapted
method
for
identification
that
could
solve
mentioned
problem.
Furthermore,
experiment
with
our
approach
800
coming
GitHub
validate
results
200
projects.
With
proposed
got
optimal
subsets
engineering
These
gave
us
low
values
at
more
than
70%
class
levels
validation
dataset.
Concurrency and Computation Practice and Experience,
Journal Year:
2025,
Volume and Issue:
37(6-8)
Published: March 13, 2025
ABSTRACT
Feature
selection
is
an
effective
tool
for
processing
data.
It
employed
to
eliminate
redundant
or
irrelevant
features
and
select
optimal
feature
subsets
improve
the
performance
of
learning
models.
The
gradient‐based
optimizer
(GBO)
received
extensive
attention
in
solving
different
optimization
problems,
which
have
gradient
search
rule
(GSR)
local
escaping
operation
(LEO).
However,
when
addressing
complex
GBO
exhibits
deficiencies
balancing
global
exploration
exploitation,
tends
converge
optima.
This
article
presents
a
modified
version
GBO,
named
FWZGBO,
problems.
Firstly,
inspired
by
iterative
method
its
theory,
we
propose
enhanced
strategy
significantly
accelerating
capability
GSR.
utilizes
fourth‐order
perform
corresponding
function
second‐order
Newton's
method.
Secondly,
suggest
refraction
approach
with
Gaussian
distribution
help
algorithm
escape
from
optima
enhance
population
diversity.
Thirdly,
this
work
devises
new
adaptive
weight
based
on
cosine
both
GSR
LEO
attain
harmonious
balance
between
exploitation.
To
validate
FWZGBO
algorithm,
28
benchmark
functions
20
well‐known
datasets
are
tested
compared
14
algorithms.
experimental
results
show
that
superior
Meanwhile,
effectiveness
validated
using
Friedman
test
post‐hoc
test.
Biomimetics,
Journal Year:
2025,
Volume and Issue:
10(4), P. 226 - 226
Published: April 4, 2025
An
improved
black-winged
kite
algorithm
with
multiple
strategies
(BKAIM)
is
proposed
in
this
paper
to
address
two
critical
limitations
the
original
optimization
(BKA):
restricted
search
capability
caused
by
low-quality
initial
population
and
reduced
diversity
resulting
from
blind
following
behavior
during
migration
phase.
Our
enhancement
implements
three
strategic
modifications
across
different
stages.
During
initialization,
an
opposition-based
learning
strategy
was
incorporated
generate
a
higher-quality
population.
For
phase,
differential
mutation
integrated
facilitate
information
exchange
among
members,
mitigate
tendency
of
leader-following
behavior,
enhance
convergence
precision,
achieve
optimal
balance
between
exploration
exploitation
capabilities.
Regarding
boundary
handling,
conventional
absorption
method
replaced
random
approach
increase
subsequently
improve
algorithm’s
Comprehensive
testing
conducted
on
four
benchmark
function
sets
(CEC2017,
CEC2019,
CEC2021,
CEC2022)
validate
effectiveness
algorithm.
Detailed
analysis
Wilcoxon
rank-sum
test
comparisons
other
algorithms
demonstrated
BKAIM’s
superior
performance
robustness.
Furthermore,
support
vector
machine
(SVM)
model
optimized
BKAIM
for
grade
identification
Dendrobium
huoshanense
based
near-infrared
spectral
data,
thereby
confirming
its
practical
applications.
Frontiers in Big Data,
Journal Year:
2025,
Volume and Issue:
7
Published: Jan. 14, 2025
Predictions
of
student
performance
are
important
to
the
education
system
as
a
whole,
helping
students
know
how
their
learning
is
changing
and
adjusting
teachers'
school
policymakers'
plans
for
future
growth.
However,
selecting
meaningful
features
from
huge
amount
educational
data
challenging,
so
dimensionality
achievement
needs
be
reduced.
Based
on
this
motivation,
paper
proposes
an
improved
Binary
Snake
Optimizer
(MBSO)
wrapped
feature
selection
model,
taking
Mat
Por
in
UCI
database
example,
comparing
MBSO
model
with
other
methods,
able
select
strong
correlation
average
number
selected
reaches
minimum
7.90
7.10,
which
greatly
reduces
complexity
prediction.
In
addition,
we
propose
MDBO-BP-Adaboost
predict
students'
performance.
Firstly,
incorporates
good
point
set
initialization,
triangle
wandering
strategy
adaptive
t-distribution
obtain
Modified
Dung
Beetle
Optimization
Algorithm
(MDBO),
secondly,
it
uses
MDBO
optimize
weights
thresholds
BP
neural
network,
lastly,
optimized
network
used
weak
learner
Adaboost.
After
XGBoost,
BP,
BP-Adaboost,
DBO-BP-Adaboost
models,
experimental
results
show
that
R2
dataset
0.930
0.903,
respectively,
proves
proposed
has
better
effect
than
models
prediction
models.
