232
Figure 4. MWD data after applying noise elimination method.
Figure 5. Classified holes.
Table 1. The accuracies obtained by applying
different feature selection methods.
Feature selection method
Accuracy (%)
mRMR
63.5 ± 2.1
ReliefF
55.4 ± 3.7
BAHSIC
77.2 ± 2.9
S-BAHSIC
80.1 ± 2.5
function considers the sample similarities in a space with infinite dimensions, thus enabling
us to extract complicated dependencies.
In contrast, the results from mRMR are not accurate enough. The method uses the mutual
information criterion to investigate the pairwise relations between features and the association of each feature with the output. This criterion is only proficient at identifying linear
T~~·T" . .
0
S
10
1S
·T~~r . .
0
2
4
6
8
10
12
H
""""
""""
f:~
:
~
l
0
"
"
""""
I' I
I
q
0
"
I:f~
0
2
4
6
8
10
12
H
""""
""""
r:f-~
0
[20
"o""----:---~----:----c------c,----7--_j
""""
Hole 1
Hole 2
Plot
Predictions
O oata
"
c
@Model predictions
•
. Correct
"
0.9
"
"
Incorrect
" "
•
Predictors
"
X: lcolumn_311
0.7
D
~_310
v·
•
"
"
;;;
•
•
I
~ 0.5
•
•
0
u 0.4
3
0.3
•
••
0.2
"
0.1
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
How to Investigate features
column_311