The froth flotation technique is very essential to the economy of the world. Therefore, research is constantly needed to explore means of optimising it. A flotation process plan that lacks the appropriate dosage of reagents can lead to wastage of resources, compromised quality of the target material and financial losses. Hence, there is a need for means of predicting reagent dosages as accurately as possible.

To achieve a software artefact capable of predicting the dosage of reagents that would be needed to process a given volume of ore feed, the random forest regressor was adopted. The data gathered specifically describes the flotation process of iron ore and silica ore simultaneously and previous work on this dataset was aimed at predicting the resultant percentage concentrate from the system. In this project, however, the target variable was the Starch and Amina flow (reagents) where the p-value technique of feature selection was implemented to realise the most important variables needed to achieve the proposed solution. Also, within this project, exploratory data analysis was carried out on the data to achieve an understanding of how the variables relate to each other. Finally, the developed model will be linked to a graphical user interface window that prospective users can use to take advantage of the proposed solution within the industry.

The exploratory data analyses revealed the behaviour of the key variables needed to predict the desired outcome. The uniform increment was illustrated and averaged out to show that the values of the variables tended to stay approximately similar throughout the recorded operation. This uniformity indicates that the output from the model can be used as a guide for operation planning and the benefits of the designed solution can be accrued. Although the correlation matrix technique yielded information that was adjudged to not be useful to the project, it was worth carrying out because the lack of correlation suggests that each variable functions independent of the rest when predicting reagent flow. The random forest regressor was also found to be an effective predictive model because it delivers an excellent R2 score and this fosters confidence in the results obtained from the execution of the model.

Development of a Software Application for the Prediction of Froth Flotation Reagents Dosage