Background:
Diabetes
is
a
very
common
disease
today
and
has
acquired
worrying
focus
in
the
field
of
public
health
globally,
fact,
it
estimated
that
number
people
with
diabetes
worldwide
reached
415
million.Objective:
Propose
method
4
combined
models
based
on
Stacking
order
to
predict
diabetes.
In
addition,
web
interface
was
developed
best
model
proposed
this
study.Methods:
The
dataset
collected
from
Dataset
composed
768
patient
records
used.
data
then
pre-processed
using
Python
programming
language.
To
balance
data,
divided
into
values
an
oversampling
applied
distribute
proportionally.
Then,
divisions
were
made
balanced
cross-validation
for
training,
calibrated.
Regarding
development
base
algorithms,
7
independent
algorithms
used,
proposed,
finally
obtain
evaluation
their
respective
metrics.Results:
1A
(Logistic
regression)
Oversampling
value
Accuracy=91.50%,
Sensitivity=91.60%,
F1-Score=91.49%
Precision=
91.50%,
while
respect
metric
ROC
Curve,
Oversampling,
2A
(Random
Forest)
oversampling,
Random
Forest
(Independent)
percentage,
being
97.00%.Conclusions:
Implementing
stacking
method,
helps
make
adequate
diagnosis
Therefore,
by
improvement
prediction
observed,
surpassing
performance
Crime Prevention and Community Safety,
Journal Year:
2024,
Volume and Issue:
26(4), P. 440 - 489
Published: Nov. 20, 2024
This
research
addresses
the
potential
for
tackling
crime
volumes
and
improving
analytics
through
new
enhancement
strategies.
The
use
of
machine
learning
deep
solutions
is
increasing
in
prediction,
as
many
other
fields.
study
aims
to
strengthen
proactive
approaches
criminology
by
evaluating
effectiveness
stacking-based
ensemble
(S-BEL)
model,
which
enhance
overall
performance
combining
strengths
various
algorithms
improve
facilitate
prevention
analyzes
six
studies
leveraging
S-BEL
model
along
with
28
articles
on
seven
utilizing
models,
56
general
prediction
studies.
findings
highlight
that
stands
out
a
prominent
technique
providing
valuable
insights
law
enforcement.
Cadernos de Saúde Pública,
Journal Year:
2024,
Volume and Issue:
40(11)
Published: Jan. 1, 2024
Abstract:
Undergraduate
students
are
often
impacted
by
depression,
anxiety,
and
stress.
In
this
context,
machine
learning
may
support
mental
health
assessment.
Based
on
the
following
research
question:
“How
do
models
perform
in
detection
of
stress
among
undergraduate
students?”,
we
aimed
to
evaluate
performance
these
models.
PubMed,
Embase,
PsycINFO,
Web
Science
databases
were
searched,
aiming
at
studies
meeting
criteria:
publication
English;
targeting
university
students;
empirical
studies;
having
been
published
a
scientific
journal;
predicting
or
outcomes
via
learning.
The
certainty
evidence
was
analyzed
using
GRADE.
As
January
2024,
2,304
articles
found,
48
met
inclusion
criteria.
Different
types
data
identified,
including
behavioral,
physiological,
internet
usage,
neurocerebral,
blood
markers,
mixed
data,
as
well
demographic
mobility
data.
Among
33
that
provided
accuracy
assessment,
30
reported
values
exceeded
70%.
Accuracy
detecting
ranged
from
63%
100%,
anxiety
53.69%
97.9%,
depression
73.5%
99.1%.
Although
most
present
adequate
performance,
it
should
be
noted
47
them
only
performed
internal
validation,
which
overstate
Moreover,
GRADE
checklist
suggested
quality
very
low.
These
findings
indicate
algorithms
hold
promise
Public
Health;
however,
is
crucial
scrutinize
their
practical
applicability.
Further
invest
mainly
external
validation
Background:
Diabetes
is
a
very
common
disease
today
and
has
acquired
worrying
focus
in
the
field
of
public
health
globally,
fact,
it
estimated
that
number
people
with
diabetes
worldwide
reached
415
million.Objective:
Propose
method
4
combined
models
based
on
Stacking
order
to
predict
diabetes.
In
addition,
web
interface
was
developed
best
model
proposed
this
study.Methods:
The
dataset
collected
from
Dataset
composed
768
patient
records
used.
data
then
pre-processed
using
Python
programming
language.
To
balance
data,
divided
into
values
an
oversampling
applied
distribute
proportionally.
Then,
divisions
were
made
balanced
cross-validation
for
training,
calibrated.
Regarding
development
base
algorithms,
7
independent
algorithms
used,
proposed,
finally
obtain
evaluation
their
respective
metrics.Results:
1A
(Logistic
regression)
Oversampling
value
Accuracy=91.50%,
Sensitivity=91.60%,
F1-Score=91.49%
Precision=
91.50%,
while
respect
metric
ROC
Curve,
Oversampling,
2A
(Random
Forest)
oversampling,
Random
Forest
(Independent)
percentage,
being
97.00%.Conclusions:
Implementing
stacking
method,
helps
make
adequate
diagnosis
Therefore,
by
improvement
prediction
observed,
surpassing
performance
