6
capturing flotation dynamics, underscoring its utility as a
robust tool for optimizing flotation processes in operational
settings.
DISCUSSION
This study highlights the efficacy of AI-driven predictive
modeling in optimizing phosphate recovery. The model’s
predictive accuracy supports proactive management in flo-
tation processes, offering a consistent approach to param-
eter adjustments. While previous studies demonstrated
success in AI-driven flotation models, this paper model is
uniquely tailored for phosphate recovery, addressing the
specific operational challenges of phosphate beneficiation.
In our initial efforts to enhance machine learning mod-
els, we encountered challenges due to data quality issues.
To address this, during the data preprocessing phase, we
focused on operational relevance by filtering the dataset
according to 2023 equipment parameters. By utilizing
SME insights, we excluded all 2022 data that fell outside
these parameters. This strategy ensured the model was
trained solely on data that was valid and pertinent to cur-
rent operational conditions, emphasizing data quality over
conventional data science filtering methods. By aligning
data selection with operational parameters, we minimized
the need for imputation, which is typically required in data
science-focused data cleansing, thereby strengthening the
dataset’s integrity for model training.
As a result, the model achieved predictive accuracy,
achieving +90% without signs of overfitting, confirming
its robustness within the intended operational scope. This
outcome reflects a collaborative synergy between SME
operational expertise and data science precision, harmoniz-
ing domain knowledge with algorithmic rigor to enhance
model performance. This integration also reduced the
number of iterative model training cycles typically neces-
sary to reach high accuracy, demonstrating the value of
incorporating operational insights directly into the Data
Science lifecycle.
FUTURE WORK
Building upon the success of this predictive model, future
work could explore the development of an AI advisory tool
for flotation operations to automate recipe generation. Such
a system would leverage the model’s predictive capabilities
to take in real-time feed parameters along with a target
P2O5 grade, then produce an optimal recipe for operators
to follow, ensuring that the desired grade is consistently
achieved. By automating recipe adjustments based on cur-
rent operational data and target outcomes, this advisory
system could minimize manual interventions and improve
overall process efficiency. This direction would represent a
significant advancement toward a fully integrated, intelli-
gent flotation system, further enhancing recovery rates and
product quality.
It’s worth emphasizing that beyond the enhancements
in the flotation process, this project has played a good
role in bridging the gap between the mining operations
team and AI technology. By fostering this connection, a
solid foundation was established for future collaborations
between Mining and AI. This synergy opens new possibili-
ties for tackling other complex challenges within the min-
ing sector.
REFERENCES
[1] McCoy, J. T., &Auret, L. (2019). Machine learning
applications in mineral processing: A review. Minerals
Engineering, 132, 310–331.
[2] Ali, D., Hayat, M. B., Alagha, L., &Molatlhegi, O.
K. (2018). An evaluation of machine learning and
artificial intelligence models for predicting the flota-
tion behavior of fine high-ash coal. Advanced Powder
Technology, 29(3), 654–666.
[3] Galas, J., &Litwin, D. (2022). Machine learning
technique for recognition of flotation froth images in
a nonstable flotation process. Minerals, 12(8), 1052.
Table 1. Flotation Prediction metrics
Dataset MAE MSE RMSE MAPE
Training 0.0015 0.0004 0.0191 0.0061%
Validation 0.0038 0.0021 0.0453 0.0157%
Testing 0.0039 0.0024 0.0487 0.0157%
Figure 6. Predicted Vs. Actual P
2 O
5 values
capturing flotation dynamics, underscoring its utility as a
robust tool for optimizing flotation processes in operational
settings.
DISCUSSION
This study highlights the efficacy of AI-driven predictive
modeling in optimizing phosphate recovery. The model’s
predictive accuracy supports proactive management in flo-
tation processes, offering a consistent approach to param-
eter adjustments. While previous studies demonstrated
success in AI-driven flotation models, this paper model is
uniquely tailored for phosphate recovery, addressing the
specific operational challenges of phosphate beneficiation.
In our initial efforts to enhance machine learning mod-
els, we encountered challenges due to data quality issues.
To address this, during the data preprocessing phase, we
focused on operational relevance by filtering the dataset
according to 2023 equipment parameters. By utilizing
SME insights, we excluded all 2022 data that fell outside
these parameters. This strategy ensured the model was
trained solely on data that was valid and pertinent to cur-
rent operational conditions, emphasizing data quality over
conventional data science filtering methods. By aligning
data selection with operational parameters, we minimized
the need for imputation, which is typically required in data
science-focused data cleansing, thereby strengthening the
dataset’s integrity for model training.
As a result, the model achieved predictive accuracy,
achieving +90% without signs of overfitting, confirming
its robustness within the intended operational scope. This
outcome reflects a collaborative synergy between SME
operational expertise and data science precision, harmoniz-
ing domain knowledge with algorithmic rigor to enhance
model performance. This integration also reduced the
number of iterative model training cycles typically neces-
sary to reach high accuracy, demonstrating the value of
incorporating operational insights directly into the Data
Science lifecycle.
FUTURE WORK
Building upon the success of this predictive model, future
work could explore the development of an AI advisory tool
for flotation operations to automate recipe generation. Such
a system would leverage the model’s predictive capabilities
to take in real-time feed parameters along with a target
P2O5 grade, then produce an optimal recipe for operators
to follow, ensuring that the desired grade is consistently
achieved. By automating recipe adjustments based on cur-
rent operational data and target outcomes, this advisory
system could minimize manual interventions and improve
overall process efficiency. This direction would represent a
significant advancement toward a fully integrated, intelli-
gent flotation system, further enhancing recovery rates and
product quality.
It’s worth emphasizing that beyond the enhancements
in the flotation process, this project has played a good
role in bridging the gap between the mining operations
team and AI technology. By fostering this connection, a
solid foundation was established for future collaborations
between Mining and AI. This synergy opens new possibili-
ties for tackling other complex challenges within the min-
ing sector.
REFERENCES
[1] McCoy, J. T., &Auret, L. (2019). Machine learning
applications in mineral processing: A review. Minerals
Engineering, 132, 310–331.
[2] Ali, D., Hayat, M. B., Alagha, L., &Molatlhegi, O.
K. (2018). An evaluation of machine learning and
artificial intelligence models for predicting the flota-
tion behavior of fine high-ash coal. Advanced Powder
Technology, 29(3), 654–666.
[3] Galas, J., &Litwin, D. (2022). Machine learning
technique for recognition of flotation froth images in
a nonstable flotation process. Minerals, 12(8), 1052.
Table 1. Flotation Prediction metrics
Dataset MAE MSE RMSE MAPE
Training 0.0015 0.0004 0.0191 0.0061%
Validation 0.0038 0.0021 0.0453 0.0157%
Testing 0.0039 0.0024 0.0487 0.0157%
Figure 6. Predicted Vs. Actual P
2 O
5 values