XXXI International Mineral Processing Congress 2024 Proceedings/Washington, DC/Sep 29–Oct 3 897
PREDICTIVE MODELS PERFORMANCE
DISCUSSION
Table 2 shows the results from the predictive model assess-
ment. The correlation coefficients of the models were gen-
erally in a descending order of GPR RF ANN =SVM
LR. The highest 𝑟 values obtained by GPR model in each
instance show quantitively the strength of the linear rela-
tionship between true and predicted rougher copper recov-
ery values when GPR model was used. This results further
indicate the uniqueness of the predictive strength of GPR
model. In terms of error statistics, RMSE and MAPE indi-
cators were used in this work. The lowest RMSE values
obtained by GPR model, relative to the other investigated
predictive models, in each instance is a clear indication that
GPR predicted rougher copper recovery values are in bet-
ter agreement with true rougher recovery values. Similarly,
GPR model was the best performing model as far as MAPE
indicator is concerned. Agreeing with the correlation coef-
ficient results, random forest followed GPR with regards to
the error statistics results.
Figure 2. Sample size and corresponding computational time and training mean square
error (MSE)
Table 2. Model performance assessment using correlation co-efficient (r), root mean
square error (RMSE), mean absolute percentage error (MAPE), and variance accounted
for (VAF)
Data set Algorithm r RMSE MAPE, %VAF, %
Training SVM 0.87 0.91 0.65 74.72
GPR 0.99 0.01 0.01 99.99
ANN 0.87 0.88 0.71 76.24
LR 0.53 1.52 1.32 26.56
RF 0.97 0.35 0.23 96.14
Validation SVM 0.85 0.96 0.69 71.72
GPR 0.97 0.41 0.24 94.70
ANN 0.86 0.93 0.74 73.50
LR 0.50 1.56 1.34 24.50
RF 0.95 0.60 0.40 88.89
Testing SVM 0.86 0.93 0.68 74.87
GPR 0.97 0.41 0.24 94.65
ANN 0.86 0.92 0.74 73.32
LR 0.51 1.54 1.32 25.62
RF 0.95 0.59 0.39 89.47
PREDICTIVE MODELS PERFORMANCE
DISCUSSION
Table 2 shows the results from the predictive model assess-
ment. The correlation coefficients of the models were gen-
erally in a descending order of GPR RF ANN =SVM
LR. The highest 𝑟 values obtained by GPR model in each
instance show quantitively the strength of the linear rela-
tionship between true and predicted rougher copper recov-
ery values when GPR model was used. This results further
indicate the uniqueness of the predictive strength of GPR
model. In terms of error statistics, RMSE and MAPE indi-
cators were used in this work. The lowest RMSE values
obtained by GPR model, relative to the other investigated
predictive models, in each instance is a clear indication that
GPR predicted rougher copper recovery values are in bet-
ter agreement with true rougher recovery values. Similarly,
GPR model was the best performing model as far as MAPE
indicator is concerned. Agreeing with the correlation coef-
ficient results, random forest followed GPR with regards to
the error statistics results.
Figure 2. Sample size and corresponding computational time and training mean square
error (MSE)
Table 2. Model performance assessment using correlation co-efficient (r), root mean
square error (RMSE), mean absolute percentage error (MAPE), and variance accounted
for (VAF)
Data set Algorithm r RMSE MAPE, %VAF, %
Training SVM 0.87 0.91 0.65 74.72
GPR 0.99 0.01 0.01 99.99
ANN 0.87 0.88 0.71 76.24
LR 0.53 1.52 1.32 26.56
RF 0.97 0.35 0.23 96.14
Validation SVM 0.85 0.96 0.69 71.72
GPR 0.97 0.41 0.24 94.70
ANN 0.86 0.93 0.74 73.50
LR 0.50 1.56 1.34 24.50
RF 0.95 0.60 0.40 88.89
Testing SVM 0.86 0.93 0.68 74.87
GPR 0.97 0.41 0.24 94.65
ANN 0.86 0.92 0.74 73.32
LR 0.51 1.54 1.32 25.62
RF 0.95 0.59 0.39 89.47