Sustainability,
Journal Year:
2023,
Volume and Issue:
15(13), P. 10660 - 10660
Published: July 6, 2023
Precise
and
efficient
air
quality
prediction
plays
a
vital
role
in
safeguarding
public
health
informing
policy-making.
Fine
particulate
matter,
specifically
PM2.5
PM10,
serves
as
crucial
indicator
for
assessing
managing
pollution
levels.
In
this
paper,
daily
concentration
model
combining
successive
variational
mode
decomposition
(SVMD)
bidirectional
long
short-term
memory
(BiLSTM)
neural
network
is
proposed.
Firstly,
SVMD
used
an
unsupervised
feature-learning
method
to
divide
data
into
intrinsic
functions
(IMFs)
extract
frequency
features
improve
trend
prediction.
Secondly,
the
BiLSTM
introduced
supervised
learning
capture
small
changes
pollutant
sequence
perform
of
decomposed
sequence.
Furthermore,
Bayesian
optimization
(BO)
algorithm
employed
identify
optimal
key
parameters
model.
Lastly,
predicted
values
are
reconstructed
generate
final
results
PM10
datasets.
The
performance
proposed
validated
using
datasets
collected
from
China
Environmental
Monitoring
Center
Tianshui,
Gansu,
Wuhan,
Hubei.
show
that
can
smooth
original
series
more
effectively
than
other
methods,
BO-BiLSTM
better
LSTM-based
models,
thereby
proving
has
excellent
feasibility
accuracy.
Frontiers in Earth Science,
Journal Year:
2025,
Volume and Issue:
13
Published: Feb. 25, 2025
Air
pollution
significantly
impacts
human
health,
making
the
development
of
effective
pollutant
concentration
assessment
methods
crucial.
This
study
introduces
a
hybrid
machine
learning
approach
to
simulate
PM
2.5
mass
using
outdoor
images,
offering
an
alternative
traditional
observation
techniques.
The
proposed
method
utilizes
convolutional
neural
network
(CNN)
extract
image
features
through
transfer
learning.
importance
these
is
then
evaluated
random
forest
(RF)
model.
In
addition,
extracted
are
combined
with
meteorological
data
(e.g.,
temperature
(TEM),
relative
humidity
(RHU),
and
sea
level
pressure
(PRS_Sea))
(hourly
concentrations
from
four
monitoring
stations)
for
complete
ensemble
empirical
mode
decomposition
adaptive
noise
(CEEMDAN)
signal
decomposition.
results
in
multiscale
signals
that
subsequently
used
model
concentrations.
demonstrate
ResNet50
training
method,
which
extracts
64
features,
yields
best
performance.
An
RF
applied
low-frequency
signal,
superimposed
trend
while
Lasso
regression
high-frequency
signal.
produces
superior
simulation
than
alone.
Notably,
feature
23,
Institute
Biological
Products
(IBP),
TEM
most
influential
characteristic
coefficients
1.409,
0.380,
0.318,
respectively.
For
signals,
5
along
Lanlian
Hotel
(LH),
significant,
values
0.170,
0.137,
0.125,
(random
model)
has
function
high
(low)
value
correction
frequency
simulation,
leading
more
accurate
results.
proposes
cost-effective
accurately
estimating
Sakarya University Journal of Computer and Information Sciences,
Journal Year:
2025,
Volume and Issue:
8(1), P. 89 - 111
Published: March 27, 2025
This
study
utilizes
air
pollution
data
from
the
Continuous
Monitoring
Center
of
Ministry
Environment,
Urbanization,
and
Climate
Change
in
Turkey
to
predict
various
pollutants
using
three
advanced
deep
learning
approaches:
LSTM
(Long
Short-Term
Memory),
CNN
(Convolutional
Neural
Network),
RNN
(Recurrent
Network).
Missing
dataset
were
imputed
K-Nearest
Neighbor
(K-NN)
algorithm
ensure
completeness.
Furthermore,
a
fusion
technique
was
applied
integrate
multiple
pollutant
enhancing
richness
reliability
input
features
for
modeling.
The
increasing
issue,
driven
by
factors
such
as
population
growth,
urbanization,
industrial
development,
is
major
environmental
concern.
evaluates
these
models
estimate
concentrations
selects
most
accurate,
RNN,
forecasting
over
next
years.
Each
prediction
assessed
performance
metrics
MAE,
RMSE,
R²
robust
model
evaluation.
Visualization
forecast
results
achieved
through
methods
like
Box
Plots,
Violin
Point
Scatter
Graphs,
making
quality
information
more
accessible
general
audiences.
In
terms
performance,
an
0.88
PM10
0.93
SO2,
while
demonstrated
0.94
0.95
SO2.
However,
emerged
accurate
model,
achieving
0.97
both
SO2
forecasts.
allows
forecasts
levels
three-year
period.
findings
indicate
that
predictive
modeling,
combined
with
visualization
techniques,
could
significantly
contribute
mitigating
future
uncertainties
enhance
comprehension
patterns
non-expert
