Supervised Training and Unsupervised Machine Learning Testing for Cobalt Dissolution from Complex Cu-Co Ores in Acid Lixiviants

Abstract

The ever-increasing application of machine learning and artificial intelligence techniques in research has had a pivotal impact on the existing body of knowledge. In the presence of huge data set available, application of machine learning probes a better handling of the data. As such, this paper discusses the philosophies involved within machine learning. The paper the discusses the supervised training and unsupervised testing stages. The study focused on a supervised model (logistic linear regression) to perform a classification predictive analysis on cobalt dissolution. It further detailed the stages of data preparation and handling of the data upon developing the model. The model's performance was validated by using performance metrics such as accuracy, precision, and recall. The model performed good obtaining an accuracy value of 75%, the precision for the "Low Yield" class is 45,45%. The recall for the "Low Yield" class is 62,5%. The precision for the "High Yield" class is 88%. The recall for the "High Yield" class is approximately 78,57%. The model showed to perform well when predicting high yield instances as compared to low yield instances. As a result, the values obtained showed that the model requires some improvements but its implementation towards the dissolution of cobalt proved to be significant as it can describe high or low yield outcomes based on specific dissolution conditions.