893
Copper Recovery Predictive Performance for Selected Machine
Learning Algorithms
B. Amankwaa-Kyeremeh
University of South Australia, UniSA STEM, Future Industries Institute, Mawson Lakes, Adelaide, Australia
C. McCamley
BHP Olympic Dam, Adelaide, Australia
R. K. Asamoah
University of South Australia, UniSA STEM, Future Industries Institute, Mawson Lakes, Adelaide, Australia
BHP Olympic Dam, Adelaide, Australia
ABSTRACT: In this paper, the predictive performance of selected machine learning algorithms (e.g., support
vector machine, Gaussian process regression, multi-layer perceptron artificial neural network, linear regression
and random forest) is investigated for copper flotation recovery. The model performance, assessed using
linear correlation (𝑟), root mean square error (RMSE), mean absolute percentage error (MAPE) and variance
accounted for (VAF), showed that Gaussian process regression model makes the most precise rougher copper
recovery prediction. The outstanding performance of the Gaussian process regression model is attributed to
its ability to adequately model the noise patterns in the copper recovery data. Links with other systems and
opportunities for further development has been discussed.
Keywords: Froth flotation, Predictive algorithms, Gaussian process regression (GPR), Machine learning, cop-
per flotation recovery, Random forest
INTRODUCTION
Copper is one of the first metals exposed to humans and has
continued to find modern commercial applications due to
its high electrical conductivity, resistance to corrosion and
its ability to easily form alloys with other metals like nickel
(cupronickel), zinc (brass) and tin (bronze) (Latorre et al.,
2019). The electrical, building, transportation and elec-
tronic industries are some of the main consumers of cop-
per produced globally (Sverdrup et al., 2014). Copper can
occur naturally in its pure state as native copper but mostly
mined and processed from primary sulphide chalcopyrite
(CuFeS2) and its oxidized states occurring near the surface
as a result of weathering (Kesler et al., 2015). With the
almost depleted oxide deposits, sulphides have remained
the major source of copper with ever-decreasing ore grades
and increasing mineralogical complexity. Most copper
mining companies are now treating large tonnages of low
grade copper ores in order to meet the rising global copper
demand which is forecasted to increase by 2–3% per year
between 2010–2050 (Kuipers et al., 2018).
Froth flotation has received widespread applications in
sulphide mineral processing, having its performance much
Copper Recovery Predictive Performance for Selected Machine
Learning Algorithms
B. Amankwaa-Kyeremeh
University of South Australia, UniSA STEM, Future Industries Institute, Mawson Lakes, Adelaide, Australia
C. McCamley
BHP Olympic Dam, Adelaide, Australia
R. K. Asamoah
University of South Australia, UniSA STEM, Future Industries Institute, Mawson Lakes, Adelaide, Australia
BHP Olympic Dam, Adelaide, Australia
ABSTRACT: In this paper, the predictive performance of selected machine learning algorithms (e.g., support
vector machine, Gaussian process regression, multi-layer perceptron artificial neural network, linear regression
and random forest) is investigated for copper flotation recovery. The model performance, assessed using
linear correlation (𝑟), root mean square error (RMSE), mean absolute percentage error (MAPE) and variance
accounted for (VAF), showed that Gaussian process regression model makes the most precise rougher copper
recovery prediction. The outstanding performance of the Gaussian process regression model is attributed to
its ability to adequately model the noise patterns in the copper recovery data. Links with other systems and
opportunities for further development has been discussed.
Keywords: Froth flotation, Predictive algorithms, Gaussian process regression (GPR), Machine learning, cop-
per flotation recovery, Random forest
INTRODUCTION
Copper is one of the first metals exposed to humans and has
continued to find modern commercial applications due to
its high electrical conductivity, resistance to corrosion and
its ability to easily form alloys with other metals like nickel
(cupronickel), zinc (brass) and tin (bronze) (Latorre et al.,
2019). The electrical, building, transportation and elec-
tronic industries are some of the main consumers of cop-
per produced globally (Sverdrup et al., 2014). Copper can
occur naturally in its pure state as native copper but mostly
mined and processed from primary sulphide chalcopyrite
(CuFeS2) and its oxidized states occurring near the surface
as a result of weathering (Kesler et al., 2015). With the
almost depleted oxide deposits, sulphides have remained
the major source of copper with ever-decreasing ore grades
and increasing mineralogical complexity. Most copper
mining companies are now treating large tonnages of low
grade copper ores in order to meet the rising global copper
demand which is forecasted to increase by 2–3% per year
between 2010–2050 (Kuipers et al., 2018).
Froth flotation has received widespread applications in
sulphide mineral processing, having its performance much