INFLUENCE OF VARIOUS TRANSFERPERFORMANCE OF ARTIFICIALNEURAL NETWORKS IN IRON ORE GRADE ESTIMATION

Various conventional and geostatistical methods are used for ore grade estimation. Inlast decade neural networks are becoming popular for this purpose. This paperfocuses on one particular feature of neural nets, namely, transfer functions andconducts a sensitivity analysis in order to determine its impact on iron ore gradeestimation in Choghart iron ore mine of Iran. This paper studies the influence ofvarious transfer functions such as logistic, tanh and Gaussian in hidden layers andlinear and nonlinear transfer functions for the output layer on the overall performanceof neural network. It is found that the combination of logistic transfer function for thehidden layer, and linear transfer function for the output layer provided highestcorrelation coefficient (R), least mean squared error (MSE) and mean absolute error(MAE). For this combination, R, MSE and MAE were found to be ۰.۷۱, ۴۰.۸ and ۴.۳,respectively.

Keywords:

ANN , Transfer function , Neural Network , Grade estimation , Iron ore , Choghart mine