5
In this study, the clusters were deemed suitable for
sample selection on the first pass of the clustering exer-
cise. However, some clusters only contained extreme grade
values and, therefore, were not used for sample selection
to prevent skewing test work results. There is potential to
improve the clustering methodology to capture the outli-
ers better using supervised machine learning techniques.
However, using supervised machine learning techniques
comes with cost and skill limitations, as training the models
requires time and specialized skills.
Validating Samples Selected
The variation in copper grade in the raw data set was
compared to the variation in copper grade in the samples
selected by comparing their histograms. The two histo-
grams (Figures 7 and 8) have a similar trend as most of
the samples chosen have a copper grade lower than 2.71%
similar to the raw drillhole data, and there is a low number
of high-grade values in both histograms.
The variation in lithology of the samples selected was
compared to the raw data to check for similarity (Tables 2
and 3). When the variation in lithology of the raw data is
similar to that of the samples selected, the samples selected
can move on to the next stage. In this study, good align-
ment between the lithology variation in the raw data and
the samples selected was achieved on the first pass. The
selected samples and the rationale for choosing them are
presented to metallurgists and geologists, and once all are
convinced the selected samples are sent for test work.
Table 2. Lithology variation of raw data
Rock Type Percentage
B3_Magnetite_carbonate Ore 22.61
B1_Biotite_calcsilicate Ore 18.41
B2_Calcsilicate_carbonate Ore 15.85
NOC 0.47
BAS 0.23
A4_Basaltic_carbonate-calcsil. Ore 4.20
A1_Graphitic Ore 4.43
CBBX 3.03
MAF 0.47
GDI 1.17
D1_Amphibole_talc_mag. Ore 3.50
A3_Tuffitic_carbonate-calcsil. Ore 11.19
O_Oxidised Ore 2.80
MSE 3.50
D2_Magnetite Ore 5.13
CLA 0.93
SCB 0.23
GAB 1.86
Table 3. Lithology variation of samples selected
Rock Type Percentage
B3_Magnetite_carbonate Ore 22.10
B1_Biotite_calcsilicate Ore 16.91
B2_Calcsilicate_carbonate Ore 12.90
NOC 0.18
BAS 0.87
A4_Basaltic_carbonate-calcsil. Ore 1.54
A1_Graphitic Ore 3.83
CBBX 1.40
MAF 0.63
GDI 0.97
D1_Amphibole_talc_mag. Ore 5.11
A3_Tuffitic_carbonate-calcsil. Ore 8.90
O_Oxidised Ore 1.07
MSE 4.16
D2_Magnetite Ore 9.79
CLA 0.55
SCB 0.93
GAB 1.26 Figure 8. Variation in the grade of main metal copper in the
samples selected.
Figure 7 .Variation in the grade of main metal cooper in the
raw dataset.
In this study, the clusters were deemed suitable for
sample selection on the first pass of the clustering exer-
cise. However, some clusters only contained extreme grade
values and, therefore, were not used for sample selection
to prevent skewing test work results. There is potential to
improve the clustering methodology to capture the outli-
ers better using supervised machine learning techniques.
However, using supervised machine learning techniques
comes with cost and skill limitations, as training the models
requires time and specialized skills.
Validating Samples Selected
The variation in copper grade in the raw data set was
compared to the variation in copper grade in the samples
selected by comparing their histograms. The two histo-
grams (Figures 7 and 8) have a similar trend as most of
the samples chosen have a copper grade lower than 2.71%
similar to the raw drillhole data, and there is a low number
of high-grade values in both histograms.
The variation in lithology of the samples selected was
compared to the raw data to check for similarity (Tables 2
and 3). When the variation in lithology of the raw data is
similar to that of the samples selected, the samples selected
can move on to the next stage. In this study, good align-
ment between the lithology variation in the raw data and
the samples selected was achieved on the first pass. The
selected samples and the rationale for choosing them are
presented to metallurgists and geologists, and once all are
convinced the selected samples are sent for test work.
Table 2. Lithology variation of raw data
Rock Type Percentage
B3_Magnetite_carbonate Ore 22.61
B1_Biotite_calcsilicate Ore 18.41
B2_Calcsilicate_carbonate Ore 15.85
NOC 0.47
BAS 0.23
A4_Basaltic_carbonate-calcsil. Ore 4.20
A1_Graphitic Ore 4.43
CBBX 3.03
MAF 0.47
GDI 1.17
D1_Amphibole_talc_mag. Ore 3.50
A3_Tuffitic_carbonate-calcsil. Ore 11.19
O_Oxidised Ore 2.80
MSE 3.50
D2_Magnetite Ore 5.13
CLA 0.93
SCB 0.23
GAB 1.86
Table 3. Lithology variation of samples selected
Rock Type Percentage
B3_Magnetite_carbonate Ore 22.10
B1_Biotite_calcsilicate Ore 16.91
B2_Calcsilicate_carbonate Ore 12.90
NOC 0.18
BAS 0.87
A4_Basaltic_carbonate-calcsil. Ore 1.54
A1_Graphitic Ore 3.83
CBBX 1.40
MAF 0.63
GDI 0.97
D1_Amphibole_talc_mag. Ore 5.11
A3_Tuffitic_carbonate-calcsil. Ore 8.90
O_Oxidised Ore 1.07
MSE 4.16
D2_Magnetite Ore 9.79
CLA 0.55
SCB 0.93
GAB 1.26 Figure 8. Variation in the grade of main metal copper in the
samples selected.
Figure 7 .Variation in the grade of main metal cooper in the
raw dataset.