Heliyon,
Год журнала:
2024,
Номер
10(16), С. e35595 - e35595
Опубликована: Авг. 1, 2024
Providing
accurate
prediction
of
the
severity
traffic
collisions
is
vital
to
improve
efficiency
emergencies
and
reduce
casualties,
accordingly
improving
safety
reducing
congestion.
However,
issue
both
predictive
accuracy
model
interpretability
predicted
outcomes
has
remained
a
persistent
challenge.
We
propose
Random
Forest
optimized
by
Meta-heuristic
algorithm
framework
that
integrates
spatiotemporal
characteristics
crashes.
Through
analysis
motor
vehicle
crash
data
on
interstate
highways
within
United
States
in
2020,
we
compared
various
ensemble
models
single-classification
models.
The
results
show
(RF)
Crown
Porcupine
Optimizer
(CPO)
best
results,
accuracy,
recall,
f1
score,
precision
can
reach
more
than
90
%.
found
factors
such
as
Temperature
Weather
are
closely
related
Closely
indicators
were
analyzed
interpretatively
using
geographic
information
system
(GIS)
based
characteristic
importance
ranking
results.
enables
crashes
discovers
important
leading
with
an
explanation.
study
proposes
some
areas
consideration
should
be
given
adding
measures
nighttime
lighting
devices
fatigue
driving
alert
ensure
safe
driving.
It
offers
references
for
policymakers
address
management
urban
development
issues.
Accurate
precipitation
prediction
is
very
important
for
meteorological
disaster
prevention,
water
resources
management,
and
agricultural
decision
making.
To
improve
the
accuracy
of
prediction,
a
hybrid
model
based
on
variational
mode
decomposition
(VMD),
crested
porcupine
optimization
algorithm
(CPO),
long
short-term
memory
(LSTM)
proposed
in
this
paper.
The
first
uses
VMD
to
decompose
time
series
into
intrinsic
functions
different
frequencies
capture
multi-scale
characteristics
data.
Then,
CPO
used
optimize
LSTM
adaptive
parameters
global
search
ability
robustness
model.
Finally,
decomposed
component
input
network
learn
spatiotemporal
dependence
relationship
long-term
prediction.
experimental
results
show
that
compared
with
traditional
model,
CPO-LSTM
VMD-LSTM
achieves
better
performance
many
evaluation
indices
effectively
improves
application
can
provide
an
effective
tool
fields
meteorology
as
well
new
ideas
related
research.
Heliyon,
Год журнала:
2024,
Номер
10(16), С. e35595 - e35595
Опубликована: Авг. 1, 2024
Providing
accurate
prediction
of
the
severity
traffic
collisions
is
vital
to
improve
efficiency
emergencies
and
reduce
casualties,
accordingly
improving
safety
reducing
congestion.
However,
issue
both
predictive
accuracy
model
interpretability
predicted
outcomes
has
remained
a
persistent
challenge.
We
propose
Random
Forest
optimized
by
Meta-heuristic
algorithm
framework
that
integrates
spatiotemporal
characteristics
crashes.
Through
analysis
motor
vehicle
crash
data
on
interstate
highways
within
United
States
in
2020,
we
compared
various
ensemble
models
single-classification
models.
The
results
show
(RF)
Crown
Porcupine
Optimizer
(CPO)
best
results,
accuracy,
recall,
f1
score,
precision
can
reach
more
than
90
%.
found
factors
such
as
Temperature
Weather
are
closely
related
Closely
indicators
were
analyzed
interpretatively
using
geographic
information
system
(GIS)
based
characteristic
importance
ranking
results.
enables
crashes
discovers
important
leading
with
an
explanation.
study
proposes
some
areas
consideration
should
be
given
adding
measures
nighttime
lighting
devices
fatigue
driving
alert
ensure
safe
driving.
It
offers
references
for
policymakers
address
management
urban
development
issues.
