<s>
Applications	O
of	O
machine	B-General_Concept
learning	I-General_Concept
in	I-General_Concept
earth	I-General_Concept
sciences	I-General_Concept
include	O
geological	O
mapping	O
,	O
gas	O
leakage	O
detection	O
and	O
geological	O
features	O
identification	O
.	O
</s>
<s>
Machine	O
learning	O
(	O
ML	O
)	O
is	O
a	O
type	O
of	O
artificial	B-Application
intelligence	I-Application
(	O
AI	B-Application
)	O
that	O
enables	O
computer	O
systems	O
to	O
classify	O
,	O
cluster	O
,	O
identify	O
and	O
analyze	O
vast	O
and	O
complex	O
sets	O
of	O
data	O
while	O
eliminating	O
the	O
need	O
for	O
explicit	O
instructions	O
and	O
programming	O
.	O
</s>
<s>
For	O
example	O
,	O
convolutional	B-Architecture
neural	I-Architecture
networks	I-Architecture
(	O
CNN	B-Architecture
)	O
are	O
good	O
at	O
interpreting	O
images	O
,	O
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
(	O
ANN	O
)	O
perform	O
well	O
in	O
soil	O
classification	O
but	O
more	O
computationally	O
expensive	O
to	O
train	O
than	O
support-vector	B-Algorithm
machine	I-Algorithm
(	O
SVM	B-Algorithm
)	O
learning	O
.	O
</s>
<s>
The	O
application	O
of	O
machine	O
learning	O
has	O
been	O
popular	O
in	O
recent	O
decades	O
,	O
as	O
the	O
development	O
of	O
other	O
technologies	O
such	O
as	O
unmanned	O
aerial	O
vehicles	O
(	O
UAVs	O
)	O
,	O
ultra-high	O
resolution	O
remote	B-General_Concept
sensing	I-General_Concept
technology	O
and	O
high-performance	O
computing	O
units	O
lead	O
to	O
the	O
availability	O
of	O
large	O
high-quality	O
datasets	O
and	O
more	O
advanced	O
algorithms	O
.	O
</s>
<s>
Applying	O
remote	B-General_Concept
sensing	I-General_Concept
with	O
machine	O
learning	O
approaches	O
provides	O
an	O
alternative	O
way	O
for	O
rapid	O
mapping	O
without	O
the	O
need	O
of	O
manually	O
mapping	O
in	O
the	O
unreachable	O
areas	O
.	O
</s>
<s>
Incorporation	O
of	O
remote	B-General_Concept
sensing	I-General_Concept
and	O
machine	O
learning	O
approaches	O
can	O
provide	O
an	O
alternative	O
solution	O
to	O
eliminate	O
some	O
field	O
mapping	O
needs	O
.	O
</s>
<s>
For	O
example	O
,	O
the	O
lithological	O
mapping	O
of	O
gold-bearing	O
granite-greenstone	O
rocks	O
in	O
Hutti	O
,	O
India	O
with	O
AVIRIS-NG	O
hyperspectral	B-Application
data	O
,	O
shows	O
more	O
than	O
10%	O
difference	O
in	O
overall	O
accuracy	O
between	O
using	O
Support	B-Algorithm
Vector	I-Algorithm
Machine	I-Algorithm
(	O
SVM	B-Algorithm
)	O
and	O
random	B-Algorithm
forest	I-Algorithm
.	O
</s>
<s>
For	O
example	O
,	O
although	O
the	O
support-vector	B-Algorithm
machine	I-Algorithm
(	O
SVM	B-Algorithm
)	O
yielded	O
the	O
best	O
result	O
in	O
landslide	O
susceptibility	O
assessment	O
accuracy	O
,	O
the	O
result	O
cannot	O
be	O
rewritten	O
in	O
the	O
form	O
of	O
expert	O
rules	O
that	O
explain	O
how	O
and	O
why	O
an	O
area	O
was	O
classified	O
as	O
that	O
specific	O
class	O
.	O
</s>
<s>
In	O
contrast	O
,	O
the	O
decision	B-Algorithm
tree	I-Algorithm
has	O
a	O
transparent	O
model	O
that	O
can	O
be	O
understood	O
easily	O
,	O
and	O
the	O
user	O
can	O
observe	O
and	O
fix	O
the	O
bias	O
if	O
any	O
present	O
in	O
the	O
model	O
.	O
</s>
<s>
If	O
the	O
computational	O
power	O
is	O
a	O
concern	O
,	O
a	O
more	O
computationally	O
demanding	O
learning	O
method	O
such	O
as	O
artificial	B-Architecture
neural	I-Architecture
network	I-Architecture
is	O
less	O
preferred	O
despite	O
the	O
fact	O
that	O
artificial	B-Architecture
neural	I-Architecture
network	I-Architecture
may	O
slightly	O
outperform	O
other	O
algorithms	O
,	O
such	O
as	O
in	O
soil	O
classification	O
.	O
</s>
<s>
Geological/	O
Lithological	O
Mapping	O
and	O
Mineral	O
Prospectivity	O
Mapping	O
can	O
be	O
carried	O
out	O
by	O
processing	O
the	O
data	O
with	O
machine-learning	O
techniques	O
with	O
the	O
input	O
of	O
spectral	O
imagery	O
obtained	O
from	O
remote	B-General_Concept
sensing	I-General_Concept
and	O
geophysical	O
data	O
.	O
</s>
<s>
Random	B-Algorithm
Forest	I-Algorithm
and	O
Support	B-Algorithm
Vector	I-Algorithm
Machine	I-Algorithm
(	O
SVM	B-Algorithm
)	O
etc	O
are	O
common	O
algorithms	O
being	O
used	O
with	O
remote	O
sensed	O
geophysical	O
data	O
,	O
while	O
Simple	O
Linear	O
Iterative	O
Clustering-Convolutional	O
Neural	O
Network	O
(	O
SLIC-CNN	O
)	O
and	O
Convolutional	B-Architecture
Neural	I-Architecture
Networks	I-Architecture
(	O
CNN	B-Architecture
)	O
etc	O
are	O
commonly	O
applied	O
while	O
dealing	O
with	O
aerial	O
photos	O
and	O
images	O
.	O
</s>
<s>
Large	O
scale	O
mapping	O
can	O
be	O
carried	O
out	O
with	O
geophysical	O
data	O
from	O
airborne	O
and	O
satellite	O
remote	B-General_Concept
sensing	I-General_Concept
geophysical	O
data	O
,	O
and	O
smaller-scale	O
mapping	O
can	O
be	O
carried	O
out	O
with	O
images	O
from	O
Unmanned	O
Aerial	O
Vehicle	O
(	O
UAV	O
)	O
for	O
higher	O
resolution	O
.	O
</s>
<s>
Vegetation	O
cover	O
is	O
one	O
of	O
the	O
major	O
obstacles	O
for	O
geological	O
mapping	O
with	O
remote	B-General_Concept
sensing	I-General_Concept
,	O
as	O
reported	O
in	O
various	O
research	O
,	O
both	O
in	O
large-scale	O
and	O
small-scale	O
mapping	O
.	O
</s>
<s>
+Examples	O
of	O
application	O
in	O
Geological/	O
Lithological	O
Mapping	O
and	O
Mineral	O
Prospectivity	O
MappingObjectiveInput	O
datasetLocationMachine	O
Learning	O
Algorithms	O
(	O
MLAs	O
)	O
PerformanceLithological	O
Mapping	O
of	O
Gold-bearing	O
granite-greenstone	O
rocksAVIRIS-NG	O
hyperspectral	B-Application
dataHutti	O
,	O
IndiaLinear	O
Discriminant	B-General_Concept
Analysis	I-General_Concept
(	O
LDA	O
)	O
,	O
</s>
<s>
Random	B-Algorithm
Forest	I-Algorithm
,	O
</s>
<s>
Support	B-Algorithm
Vector	I-Algorithm
Machine	I-Algorithm
(	O
SVM	B-Algorithm
)	O
Support	B-Algorithm
Vector	I-Algorithm
Machine	I-Algorithm
(	O
SVM	B-Algorithm
)	O
outperforms	O
the	O
other	O
Machine	B-Algorithm
Learning	I-Algorithm
Algorithms	I-Algorithm
(	O
MLAs	O
)	O
Lithological	O
Mapping	O
in	O
the	O
Tropical	O
RainforestMagnetic	O
Vector	O
Inversion	O
,	O
</s>
<s>
Shuttle	B-Algorithm
Radar	I-Algorithm
Topography	I-Algorithm
Mission	I-Algorithm
(	O
SRTM	B-Algorithm
)	O
,	O
</s>
<s>
(	O
2	O
)	O
Map	O
generated	O
with	O
remote	B-General_Concept
sensing	I-General_Concept
data	O
and	O
spatial	O
constraints	O
has	O
a	O
78.7	O
%	O
accuracy	O
but	O
no	O
new	O
possible	O
lithological	O
units	O
are	O
identifiedGeological	O
Mapping	O
for	O
mineral	O
explorationRadford	O
,	O
D	O
.	O
D.	O
,	O
Cracknell	O
,	O
M	O
.	O
J.	O
,	O
Roach	O
,	O
M	O
.	O
J.	O
,	O
&	O
Cumming	O
,	O
G	O
.	O
V	O
.	O
(	O
2018	O
)	O
.	O
</s>
<s>
Geological	O
mapping	O
in	O
western	O
Tasmania	O
using	O
radar	O
and	O
random	B-Algorithm
forests	I-Algorithm
.	O
</s>
<s>
IEEE	O
Journal	O
of	O
Selected	O
Topics	O
in	O
Applied	O
Earth	O
Observations	O
and	O
Remote	B-General_Concept
Sensing	I-General_Concept
,	O
11(9 )	O
,	O
3075-3087.Airborne	O
polarimetric	O
Terrain	O
Observation	O
with	O
Progressive	O
Scans	O
SAR	B-Application
(	O
TopSAR	O
)	O
,	O
</s>
<s>
geophysical	O
dataWestern	O
TasmaniaRandom	O
ForestLow	O
reliability	O
of	O
TopSAR	O
for	O
geological	O
mapping	O
,	O
but	O
accurate	O
with	O
geophysical	O
data.Geological	O
and	O
Mineralogical	O
mappingMultispectral	O
and	O
hyperspectral	B-Application
satellite	O
dataCentral	O
Jebilet	O
,	O
</s>
<s>
MoroccoSupport	O
Vector	O
Machine	O
(	O
SVM	B-Algorithm
)	O
The	O
accuracy	O
of	O
using	O
hyperspectral	B-Application
data	O
for	O
classifying	O
is	O
slightly	O
higher	O
than	O
that	O
using	O
multispectral	O
data	O
,	O
obtaining	O
93.05	O
%	O
and	O
89.24	O
%	O
respectively	O
,	O
showing	O
that	O
machine	O
learning	O
is	O
a	O
reliable	O
tool	O
for	O
mineral	O
exploration.Integrating	O
Multigeophysical	O
Data	O
into	O
a	O
Cluster	O
MapWang	O
,	O
Y.	O
,	O
Ksienzyk	O
,	O
A	O
.	O
K.	O
,	O
Liu	O
,	O
M.	O
,	O
&	O
Brönner	O
,	O
M	O
.	O
(	O
2021	O
)	O
.	O
</s>
<s>
ChinaSimple	O
Linear	O
Iterative	O
Clustering-Convolutional	O
Neural	O
Network	O
(	O
SLIC-CNN	O
)	O
The	O
result	O
is	O
satisfactory	O
in	O
mapping	O
major	O
geological	O
units	O
but	O
showed	O
poor	O
performance	O
in	O
mapping	O
pegmatites	O
,	O
fine-grained	O
rocks	O
and	O
dykes	O
.	O
</s>
<s>
CanadaConvolutional	O
Neural	B-Architecture
Networks	I-Architecture
(	O
CNN	B-Architecture
)	O
,	O
</s>
<s>
Random	O
ForestThe	O
resulting	O
accuracy	O
of	O
CNN	B-Architecture
was	O
76%	O
in	O
the	O
locally	O
trained	O
area	O
,	O
while	O
68%	O
for	O
an	O
independent	O
test	O
area	O
.	O
</s>
<s>
The	O
CNN	B-Architecture
achieved	O
a	O
slightly	O
higher	O
accuracy	O
of	O
4%	O
than	O
the	O
Random	B-Algorithm
Forest	I-Algorithm
.	O
</s>
<s>
Input	O
dataset	O
for	O
machine	B-Algorithm
learning	I-Algorithm
algorithms	I-Algorithm
usually	O
includes	O
topographic	O
information	O
,	O
lithological	O
information	O
,	O
satellite	O
images	O
,	O
etc	O
.	O
</s>
<s>
SerbiaSupport	O
Vector	O
Machine	O
(	O
SVM	B-Algorithm
)	O
,	O
</s>
<s>
Decision	B-Algorithm
Trees	I-Algorithm
,	O
</s>
<s>
Logistic	O
RegressionSupport	O
Vector	O
Machine	O
(	O
SVM	B-Algorithm
)	O
outperforms	O
the	O
othersLandslide	O
Susceptibility	O
MappingASTER	O
satellite-based	O
geomorphic	O
data	O
,	O
</s>
<s>
JapanArtificial	O
Neural	B-Architecture
Network	I-Architecture
(	O
ANN	O
)	O
Accuracy	O
greater	O
than	O
90%	O
for	O
determining	O
the	O
probability	O
of	O
landslide.Landslide	O
Susceptibility	O
Zonation	O
through	O
ratingsChauhan	O
,	O
S.	O
,	O
Sharma	O
,	O
M.	O
,	O
Arora	O
,	O
M	O
.	O
K.	O
,	O
&	O
Gupta	O
,	O
N	O
.	O
K	O
.	O
(	O
2010	O
)	O
.	O
</s>
<s>
Landslide	O
susceptibility	O
zonation	O
through	O
ratings	O
derived	O
from	O
artificial	B-Architecture
neural	I-Architecture
network	I-Architecture
.	O
</s>
<s>
IndiaArtificial	O
Neural	B-Architecture
Network	I-Architecture
(	O
ANN	O
)	O
The	O
AUC	O
of	O
this	O
approach	O
reaches	O
0.88	O
.	O
</s>
<s>
MalaysiaArtificial	O
Neural	B-Architecture
Network	I-Architecture
(	O
ANN	O
)	O
The	O
approach	O
achieved	O
82.92	O
%	O
accuracy	O
of	O
prediction	O
.	O
</s>
<s>
Rock	O
fractures	O
can	O
be	O
recognized	O
automatically	O
by	O
machine	O
learning	O
through	O
photogrammetric	B-Application
analysis	O
even	O
with	O
the	O
presence	O
of	O
interfering	O
objects	O
,	O
for	O
example	O
,	O
foliation	O
,	O
rod-shaped	O
vegetation	O
,	O
etc	O
.	O
</s>
<s>
In	O
machine	O
training	O
for	O
classifying	O
images	O
,	O
data	B-General_Concept
augmentation	I-General_Concept
is	O
a	O
common	O
practice	O
to	O
avoid	O
overfitting	B-Error_Name
and	O
increase	O
the	O
training	O
dataset	O
.	O
</s>
<s>
Data	B-General_Concept
augmentation	I-General_Concept
was	O
then	O
carried	O
out	O
and	O
the	O
training	O
dataset	O
was	O
increased	O
to	O
8704	O
images	O
by	O
flip	O
and	O
random	O
crop	O
.	O
</s>
<s>
KoreaConvolutional	O
Neural	B-Architecture
Network	I-Architecture
(	O
CNN	B-Architecture
)	O
The	O
approach	O
was	O
able	O
to	O
recognize	O
the	O
rock	O
fractures	O
accurately	O
in	O
most	O
cases	O
.	O
</s>
<s>
Carbon	O
dioxide	O
leakage	O
from	O
a	O
geologic	O
sequestration	O
site	O
can	O
be	O
detected	O
indirectly	O
by	O
planet	O
stress	O
response	O
with	O
the	O
aid	O
of	O
remote	B-General_Concept
sensing	I-General_Concept
and	O
an	O
unsupervised	B-General_Concept
clustering	I-General_Concept
algorithm	I-General_Concept
(	O
Iterative	O
Self-Organizing	O
Data	O
Analysis	O
Technique	O
(	O
ISODATA	O
)	O
method	O
)	O
.	O
</s>
<s>
The	O
hyperspectral	B-Application
images	I-Application
are	O
processed	O
by	O
the	O
unsupervised	O
algorithm	O
clustering	O
pixels	O
with	O
similar	O
plant	O
responses	O
.	O
</s>
<s>
The	O
hyperspectral	B-Application
information	O
in	O
areas	O
with	O
known	O
CO2	O
leakage	O
was	O
extracted	O
so	O
that	O
areas	O
with	O
CO2	O
leakage	O
can	O
be	O
matched	O
with	O
the	O
clustered	O
pixels	O
with	O
spectral	O
anomalies	O
.	O
</s>
<s>
+Examples	O
of	O
application	O
in	O
Carbon	O
Dioxide	O
Leakage	O
DetectionObjectiveInput	O
datasetLocationMachine	O
Learning	O
Algorithms	O
(	O
MLAs	O
)	O
PerformanceDetection	O
of	O
CO2	O
leak	O
from	O
a	O
geologic	O
sequestration	O
siteAerial	O
hyperspectral	B-Application
imageryThe	O
Zero	O
Emissions	O
Research	O
and	O
Technology	O
(	O
ZERT	O
)	O
,	O
</s>
<s>
+Examples	O
of	O
application	O
in	O
Quantification	O
of	O
Water	O
InflowObjectiveInput	O
datasetLocationMachine	O
Learning	O
Algorithms	O
(	O
MLAs	O
)	O
PerformanceQuantification	O
of	O
water	O
inflow	O
in	O
rock	O
tunnel	O
facesImages	O
of	O
water	O
inflow	O
-Convolutional	O
Neural	O
Network	O
(	O
CNN	B-Architecture
)	O
The	O
approach	O
achieved	O
an	O
average	O
accuracy	O
of	O
93.01	O
%	O
.	O
</s>
<s>
The	O
classification	O
part	O
can	O
be	O
carried	O
out	O
by	O
Decision	B-Algorithm
Trees	I-Algorithm
(	O
DT	O
)	O
,	O
Artificial	B-Architecture
Neural	I-Architecture
Network	I-Architecture
(	O
ANN	O
)	O
,	O
or	O
Support	B-Algorithm
Vector	I-Algorithm
Machine	I-Algorithm
(	O
SVM	B-Algorithm
)	O
.	O
</s>
<s>
While	O
comparing	O
the	O
three	O
algorithms	O
,	O
it	O
is	O
demonstrated	O
that	O
the	O
Artificial	B-Architecture
Neural	I-Architecture
Network	I-Architecture
(	O
ANN	O
)	O
performed	O
the	O
best	O
in	O
classifying	O
humous	O
clay	O
and	O
peat	O
,	O
while	O
the	O
Decision	B-Algorithm
Trees	I-Algorithm
performed	O
the	O
best	O
in	O
classifying	O
clayey	O
peat	O
.	O
</s>
<s>
Artificial	B-Architecture
Neural	I-Architecture
Network	I-Architecture
(	O
ANN	O
)	O
,	O
</s>
<s>
Support	O
Vector	O
MachineThe	O
Artificial	B-Architecture
Neural	I-Architecture
Network	I-Architecture
(	O
ANN	O
)	O
outperformed	O
the	O
others	O
in	O
classifying	O
humous	O
clay	O
and	O
peat	O
,	O
while	O
the	O
Decision	B-Algorithm
Trees	I-Algorithm
outperformed	O
the	O
others	O
in	O
classifying	O
clayey	O
peat	O
.	O
</s>
<s>
Support	B-Algorithm
Vector	I-Algorithm
Machine	I-Algorithm
gave	O
the	O
poorest	O
performance	O
among	O
the	O
three	O
.	O
</s>
<s>
Exposed	O
geological	O
structures	O
like	O
anticline	O
,	O
ripple	O
marks	O
,	O
xenolith	O
,	O
scratch	O
,	O
ptygmatic	O
folds	O
,	O
fault	O
,	O
concretion	O
,	O
mudcracks	O
,	O
gneissose	O
,	O
boudin	O
,	O
basalt	O
columns	O
and	O
dike	O
can	O
be	O
identified	O
automatically	O
with	O
a	O
deep	B-Algorithm
learning	I-Algorithm
model	O
.	O
</s>
<s>
Research	O
demonstrated	O
that	O
Three-layer	O
Convolutional	B-Architecture
Neural	I-Architecture
Network	I-Architecture
(	O
CNN	B-Architecture
)	O
and	O
Transfer	B-General_Concept
Learning	I-General_Concept
have	O
great	O
accuracy	O
of	O
about	O
80%	O
and	O
90%	O
respectively	O
,	O
while	O
others	O
like	O
K-nearest	B-General_Concept
neighbors	I-General_Concept
(	O
KNN	O
)	O
,	O
Artificial	B-Architecture
Neural	I-Architecture
Network	I-Architecture
(	O
ANN	O
)	O
and	O
Extreme	O
Gradient	O
Boosting	O
(	O
XGBoost	O
)	O
have	O
low	O
accuracies	O
,	O
ranges	O
from	O
10%	O
-	O
30%	O
.	O
</s>
<s>
Artificial	B-Architecture
Neural	I-Architecture
Network	I-Architecture
(	O
ANN	O
)	O
,	O
</s>
<s>
Three-layer	O
Convolutional	B-Architecture
Neural	I-Architecture
Network	I-Architecture
(	O
CNN	B-Architecture
)	O
,	O
</s>
<s>
Transfer	O
LearningThree-layer	O
Convolutional	B-Architecture
Neural	I-Architecture
Network	I-Architecture
(	O
CNN	B-Architecture
)	O
and	O
Transfer	B-General_Concept
Learning	I-General_Concept
reached	O
accuracies	O
up	O
to	O
about	O
80%	O
and	O
90%	O
respectively	O
,	O
while	O
others	O
were	O
relatively	O
low	O
,	O
ranges	O
from	O
about	O
10%	O
to	O
30%	O
.	O
</s>
<s>
The	O
method	O
consists	O
of	O
two	O
parts	O
,	O
the	O
first	O
part	O
is	O
unsupervised	B-General_Concept
learning	I-General_Concept
with	O
Generative	B-Algorithm
Adversarial	I-Algorithm
Network	I-Algorithm
(	O
GAN	O
)	O
to	O
learn	O
and	O
extract	O
features	O
of	O
first	O
arrival	O
P-waves	B-Algorithm
,	O
and	O
Random	B-Algorithm
Forest	I-Algorithm
to	O
discriminate	O
P-waves	B-Algorithm
.	O
</s>
<s>
The	O
approach	O
achieved	O
99.2	O
%	O
in	O
recognizing	O
P-waves	B-Algorithm
and	O
can	O
avoid	O
false	O
triggers	O
by	O
noise	O
signals	O
with	O
98.4	O
%	O
accuracy	O
.	O
</s>
<s>
The	O
algorithm	O
applied	O
was	O
Random	B-Algorithm
Forest	I-Algorithm
trained	O
with	O
about	O
10	O
slip	O
events	O
and	O
performed	O
excellently	O
in	O
predicting	O
the	O
remaining	O
time	O
to	O
failure	O
.	O
</s>
<s>
Random	O
ForestThe	O
approach	O
can	O
recognise	O
P	B-Algorithm
waves	I-Algorithm
with	O
99.2	O
%	O
accuracy	O
and	O
avoid	O
false	O
triggers	O
by	O
noise	O
signals	O
with	O
98.4	O
%	O
accuracy.Predicting	O
time	O
remaining	O
for	O
next	O
earthquakeContinuous	O
acoustic	O
time	O
series	O
data	O
-Random	O
ForestThe	O
R2	O
value	O
of	O
the	O
prediction	O
reached	O
0.89	O
,	O
which	O
demonstrated	O
excellent	O
performance	O
.	O
</s>
<s>
However	O
,	O
some	O
very	O
useful	O
products	O
like	O
satellite	O
remote	B-General_Concept
sensing	I-General_Concept
data	O
only	O
have	O
decades	O
of	O
data	O
since	O
the	O
1970s	O
.	O
</s>
<s>
In	O
a	O
study	O
of	O
automatic	O
classification	O
of	O
geological	O
structures	O
,	O
the	O
weakness	O
of	O
the	O
model	O
is	O
the	O
small	O
training	O
dataset	O
,	O
even	O
though	O
with	O
the	O
help	O
of	O
data	B-General_Concept
augmentation	I-General_Concept
to	O
increase	O
the	O
size	O
of	O
the	O
dataset	O
.	O
</s>
<s>
Inadequate	O
training	O
data	O
may	O
lead	O
to	O
a	O
problem	O
called	O
overfitting	B-Error_Name
.	O
</s>
<s>
Overfitting	B-Error_Name
causes	O
inaccuracies	O
in	O
machine	O
learning	O
as	O
the	O
model	O
learns	O
about	O
the	O
noise	O
and	O
undesired	O
details	O
.	O
</s>
<s>
In	O
many	O
machine	B-Algorithm
learning	I-Algorithm
algorithms	I-Algorithm
,	O
for	O
example	O
,	O
Artificial	B-Architecture
Neural	I-Architecture
Network	I-Architecture
(	O
ANN	O
)	O
,	O
it	O
is	O
considered	O
as	O
'	O
black	B-Device
box	I-Device
 '	O
approach	O
as	O
clear	O
relationships	O
and	O
descriptions	O
of	O
how	O
the	O
results	O
are	O
generated	O
in	O
the	O
hidden	O
layers	O
are	O
unknown	O
.	O
</s>
<s>
'	O
White-box	O
'	O
approach	O
such	O
as	O
decision	B-Algorithm
tree	I-Algorithm
can	O
reveal	O
the	O
algorithm	O
details	O
to	O
the	O
users	O
.	O
</s>
