Advancing Flotation Process Modeling: Bayesian vs. Sklearn Approaches for Gold Grade Prediction

Abstract

This study explores Bayesian Ridge Regression and PyMC-based probabilistic modelling to predict the cumulative grade of gold based on key operational variables in gold flotation. By integrating prior knowledge and quantifying uncertainty, the Bayesian approach enhances both interpretability and predictive accuracy. The dataset includes variables such as crusher type, particle size, power, time, head grade, and collector type. Comparative analysis reveals that PyMC outperforms traditional Sklearn models, achieving an R2 of 0.92 and an MSE of 102.37. These findings highlight the potential of Bayesian models for robust, data-driven process optimization in mineral processing. The higher cumulative gold grade observed for VSI products and PAX collector usage may be attributed to the superior liberation efficiency of VSI, which produces more angular and cleanly fractured particles, enhancing collector attachment. PAX, being a strong xanthate, shows high affinity for sulphide mineral surfaces, particularly under the flotation conditions used, thereby improving selectivity and recovery.