International Journal of Scientific Research in Engineering and Management (IJSREM)

Volume: 06 Issue: 06 | June - 2022 Impact Factor: 7.185 ISSN: 2582-3930

Iron ore Quality prediction Using Machine Learning

Chetan Kumar G S1, Kavana K M2
1Assitant Professor, Department of Master of Computer Application,UBDTCE,Davangere
2Kavana K M,PG Student, Department of Master of Computer Application,UBDTCE,Davangere

ABSTRACT:

The main goal of this project is to predict how much impurity Research Objectives
is in the ore concentrate. The% of Silica is measured in a lab
experiment it takes at least one hour for the process engineers
to have this value. As this impurity is measured every hour, if 1. To evaluate the feasibility of using machine learning
we can predict how much silica (impurity) is in the ore algorithms to predict in real-time the percentage of silica
concentrate, we can help the engineers, giving them early concentrate of froth flotation processing plant.
information to take actions (empowering!). Hence, they will
be able to take corrective actions in advance (reduce impurity,
if it is the case) and also help the environment (reducing the 2. Model selection: The project finds out which variable
amount of ore that goes to tailings as you reduce silica in the associated with iron ore extraction is statistically significant.
ore concentrate).

1.Introduction
3. Estimate: The project will propose a model to predict
The approach is simple. It aims whether we can predict the percentage of silica concentrate in froth flotation
silica concentrate without iron concentrate and approached
with simple way of developing the model with concentrate and
model without concentrate and compare the performance of 2.LITERATURE REVIEW
model using various regression metric like R^2 or MAE and Column Process DESCRIPTION OF
drawing conclusion based on the results. VARIABLES IN FORTH
PLANT
When multiple dependent variables exist in a regression Date date of the measurement
model, this task is called as multi-target regression. In this % Iron Feed % of Iron that comes
case, a multi-output regressor is employed to learn the from the iron ore that is
mapping from input features to output variables jointly. In this
being fed into the
study, multi-target regression technique is implemented for
quality prediction in a mining process to estimate the amount flotation cells
of silica and iron concentrates in the ore at the end of the % Silica Feed % of silica (impurity) that
process. In the experimental studies, different regressors that comes from the iron ore
use Random Forest, AdaBoost, k-Nearest Neighbors and that is being fed into the
Decision Tree algorithms separately in the background were flotation cells
compared to determine the best model. Coefficient of Starch Flow Starch (reagent) Flow
determination (R 2 ) measure was used as the evaluation measured in m3/h
metric. There are some studies that predict iron concentrate Amina Flow Amina (reagent) Flow
and silica concentrate separately. However, this Model measured in m3/h
provides a new contribution to the field by calculating these Ore Pulp Flow t/h
two values jointly since they have a great correlation. Ore Pulp pH pH scale from 0 to 14
Our Approaches is whether Ore Pulp Density Density scale from 1 to 3
1. % Iron Concentrate is correlated with % Silica Concentrate kg/cm³
2.Predict the % silica concentrate without using % iron Flotation Column 01 Air Air flow that goes into
concentrate . Flow the flotation cell
3. If it is correlated and we can predict both % Iron and Silica measured in Nm³/h
concentrate at same time using power of ML and DL . Flotation Column 02 Air Air flow that goes into

© 2022, IJSREM | www.ijsrem.com DOI: 10.55041/IJSREM14486 | Page 1

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 06 Issue: 06 | June - 2022 Impact Factor: 7.185 ISSN: 2582-3930

Flow the flotation cell dataset from data analytic practitioners. Data scientists
measured in Nm³/h compete to build the best model for both descriptive and
Flotation Column 03 Air Air flow that goes into predictive analytic. It however allows individual to access
Flow the flotation cell their dataset in order create models and also work with other
measured in Nm³/h data scientist to solve various real world analytics problems.
The input dataset used in developing this model has been
Flotation Column 04 Air Air flow that goes into
downloaded from Kaggle. The dataset contains design
Flow the flotation cell
characteristics of iron ore froth flotation processing plant
measured in Nm³/h which were put together within three months. This is nicely
Flotation Column 05 Air Air flow that goes into organized using common format and a standardized set of
Flow the flotation cell associate features of iron ore froth flotation system.
measured in Nm³/h
Flotation Column 06 Air Air flow that goes into Structure of Dataset
Flow the flotation cell
measured in Nm³/h The dataset contains 24 columns representing the
Flotation Column 07 Air Air flow that goes into measurements, 737,453 samples exist. The 24 columns include
the date and time of the measurement, which will not be used
Flow the flotation cell
as an input feature. The last columns of the dataset represent
measured in Nm³/h
the targets of this prediction task: the percentages of iron ore
Flotation Column 01 Froth level in the and silica concentrate, which are highly inversely correlated.
Level flotation cell measured in Our goal is to predict silica concentrate without the use of iron
mm (millimeters) concentrate. The other 21 columns will be used as features for
Flotation Column 02 Froth level in the predicting the target value. Description of each feature can be
Level flotation cell measured in found in Table above
mm (millimeters)
Flotation Column 03 Froth level in the
Level flotation cell measured in
2.2 Proposed Solution
mm (millimeters)
Flotation Column 04 Froth level in the Over the past two decades, there has been an upsurge of
Level flotation cell measured in academic research work within froth flotation process
mm (millimeters) fraternity. Though, a significant number of the plant
Flotation Column 05 Froth level in the processing problems are being successfully modelled using
Level flotation cell measured in machine learning algorithms but other unresolved issues and
mm (millimeters) impediment still remain.
Flotation Column 06 Froth level in the
Level flotation cell measured in
mm (millimeters) Random ForestRegressor
Flotation Column 07 Froth level in the
This method basically trains a number of classifying decision
Level flotation cell measured in
trees on various different subsamples. It benefits from
mm averaging mechanism to improve the predictive accuracy and
%Iron Concentrate % of Iron which to control over-fitting. Training samples are randomly selected
represents how much with replacement. The size of each new training set is the same
iron is presented in the as the original dataset. That is to say, a chosen instance is likely
end of the flotation to be chosen again and again as an element of distinct subsets.
process As input parameters, the number of trees in the algorithm and
% Silica Concentrate % of silica which maximum depth should be determined initially. The change in
represents how much their values may affect the performance and predictive power
iron is presented in the of the algorithm. Therefore, all possible parameters in the range
end of the flotation for the size of the dataset are given to the method and tested.
process The parameters leading to best results become candidates to be
2.1 Source of Data used. This method performs efficiently without causing too
much computational cost.
Kaggle is an online community for descriptive analysis and
predictive modelling. It collects variety of research fields’
© 2022, IJSREM | www.ijsrem.com DOI: 10.55041/IJSREM14486 | Page 2
 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 06 Issue: 06 | June - 2022 Impact Factor: 7.185 ISSN: 2582-3930

3.1BlockDiagram
3. THEORETICAL ANALYSIS

When multiple dependent variables exist in a regression model,

this task is called as multi-target regression. In this case, a Data set DataPre- Cheki Data
multi-output regressor is employed to learn the mapping from processing
3.2 SoftwareDesigning ng Visual
input features to output variables jointly. In this study, multi- Null ization
target regression technique is implemented for quality
prediction in a mining process to estimate the amount of silica
and iron concentrates in the ore at the end of the process.
PCA Feature Train And Spli
In this study, two inter-dependent single target regression tasks
Scaling Test Split ttin
are transformed into a multiple output regression problem for
Data g
quality prediction in a mining process.

In the previous models have been conducted to estimate silica

concentrate with or without taking iron concentrate as input
parameter. In this aspect, the problem is a single-target Deployem
regression problem. However, this study that focuses on the ent
estimation of both iron and silica concentrates simultaneously
as output variables. We compared different multi-target
regressors that use Random Forest, AdaBoost, XGBOOST
,RIDGE and Decision Tree algorithms separately in the Jupyter NotebookEnvironment
background. Coefficient of determination (R^2) metric and
MSE was used to evaluate predictive performance of the SpyderIde
regression methods for the mentioned data.
Machine LearningAlgorithms
The prediction error is defined as the difference between its
actual outcome value and its predicted outcome value.In this Python(pandas,numpy,matplotlib,seaborn,sklearn)
study, two metrics were used to compare models: - RMSE and
HTML
MAE. RMSE (root mean squared error) is calculated . This is
computed by taking the differences between the target and the Flask
actual algorithm outputs, squaring them and averaging over all
classes and internal validation samples . We developed this loan status prediction by using the Python
language which is a interpreted and high level programming
MAE (mean absolute error/deviation) is calculated as MAE language and usng the Machine Learning algorithms. for
This gives the magnitude of the average absolute error . coding we used the Jupyter Notebook environment of the
Anaconda distributions and the
Spyder,itisanintegratedscientificprogramminginthe
pythonlanguage.

For creating an user interface for the prediction we used the

Flask. It is a micro web framework written in Python. It is
classified as a microframework because it does not require
particulartools or libraries.It
hasnodatabaseabstractionlayer,formvalidation,oranyother
componentswherepre-existingthird-
partylibrariesprovidecommonfunctions,andascripting language
to create a webpage is HTML by creating the templates to use
in th functions ofthe Flask andHTML.

© 2022, IJSREM | www.ijsrem.com DOI: 10.55041/IJSREM14486 | Page 3

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 06 Issue: 06 | June - 2022 Impact Factor: 7.185 ISSN: 2582-3930

4.EXPERIMENTALINVESTIGATION 4. it can increase the thermal conductivity, change the

adhesive viscosity and increase the flame retardancy.
Below is the image of data set and it has totally 737453 data
points and 24 attributes. In this blog , we used the first 21 5. due to the fine grain size and reasonable distribution
attributes as independent variables and the last two attributes of silica fume, it can effectively reduce and eliminate
(% iron and % silica concentrate) as target variables. precipitation and stratification.

6. pure silicon powder, low content of impurities, stable

physical and chemical properties, so that the curing
5. RESULT material has good insulation properties and arc
resistance.
Dataset:-
DISADVANTAGES
In this analysis, we evaluate the predictive performance of the
aforementioned ML models. The values for RMSE, MSE, and 1. dry shrinkage.
R2 for all models are reported in Figure below. Low values for 2. it is easy to produce temperature cracks.
RMSE and MSE, and high values for R2 indicate better model 3. Silica fume requires a high amount of water and
performance, respectively. For simplicity of discussion, needs to be used with a superplasticizer.
RMSE is used as the primary metric of performance. In 4. The price of silica fume is relatively high compared
addition, both the testing and validation performance is to cement and fly ash.
reported, which facilitates the discussion on overfitting in the 5. Silica fume will increase the autogenous shrinkage of
models. These performance measures are plotted as boxplots the cement slurry, and the amount of inclusion will
to illustrate the range and variance of the error. exceed 5%, which may increase the risk of
cracking.It is easy to cause cracks in mortar and
r2_score mse rmse concrete and need concrete maintenance..
% silica 0.99417 0.002450 0.049504
concentrate 7.CONCLUSION
without using %
iron concentrate  This Project presents a simple mathematical model to
. predict the quality prediction in a mining process
0.99391 0.001925 0.043886 from the early time test results. In this study, the slica
% silica
concentrate characteristic with date is modeled by a
concentrate with
Random forest regression mathematical equation.
using % iron
Early age test data are being used in this case to get
concentrate .
reliable values of the 20 seconds silica prediction.
Herein, a simple and practical approach has been
described for prediction of quality prediction in a
6. ADVANTAGES AND DISADVANTAGES mining process and the proposed technique can be
used as a reliable tool for assessing the mining
ADVANTAGES process from quite early test results. This will help in
making quick decision at site and reduce delay in the
1. Silica is basically impurity in iron ore and execution time of large construction projects.
by predicting the impurity in ore we can help the  To predict the silica(impurity) % in the ore
engineers in the plant to take measurements in early concentrate in a less time we are building a predictive
stages of manufacturing. To help the environment by analytics system in that we are applying various
reducing the amount of ore that goes to tailing as you machine learning algorithms and find the best
reduce silica in the ore concentrate. accurate model. Here web application will be used to
2. silica fume is a kind of neutral inorganic filler with display the prediction . The web application is built
very stable physical and chemical properties. It does by using flask framework and it is integrated with
not contain crystalline water, does not participate in trained ML model.
the curing reaction, and does not affect the reaction
mechanism.

3. good infiltration for various kinds of resin, good

adsorption performance, easy to mix, no
agglomeration phenomenon.

© 2022, IJSREM | www.ijsrem.com DOI: 10.55041/IJSREM14486 | Page 4

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 06 Issue: 06 | June - 2022 Impact Factor: 7.185 ISSN: 2582-3930

8. REFERENCES Prediction in a Mining Process. Retrieved September

7, 2019, from Kaggle.com website:
 Breiman, L., &Schapire, R. (2001). Random Forests. https://www.kaggle.com/edumagalhaes/quality-
IEEE, 45, 5–32. Retrieved from prediction-in-a-mining-process
https://www4.stat.ncsu.edu/~lu/ST7901/reading%20  Li, C., Sun, H., Bai, J., & Li, L. (2010). Innovative
materials/Breiman2001_Article_Random Forests.pdf. methodology for comprehensive utilization of iron
 https://ieeexplore.ieee.org/abstract/document/8907120 ore tailings. Journal of Hazardous Materials
.
 Eduardo Magalhães Oliveira. (2017). Quality

© 2022, IJSREM | www.ijsrem.com DOI: 10.55041/IJSREM14486 | Page 5