Comparing the predictive performance of neural networks with ordinary kriging in a bauxite deposit

Abstract

This paper investigates the performance of neural network and ordinary kriging techniques in estimating five variables: accumulated Al₂O₃, accumulated SiO₂, percentage Al₂O₃, percentage SiO₂, and thickness of a bauxite deposit in India. The assay values obtained from exploratory boreholes were compiled according to the geological composition of the deposit. Genetic algorithms were used to divide the dataset into model development and evaluation subsets, ensuring that the modelling subset and evaluation subset were similar and that performance evaluation was valid. The results indicate that neural networks and ordinary kriging performed equally well in this deposit, except for the variable accumulated Al₂O₃. However, the coefficients of determination (R²) of predictions were not very good.