Document Type : Original Research Paper

Authors

• Tapan Dey ¹
• Gopinath Samanta ²

¹ Research Scholar at NIT Rourkela

² Senior Technologist, Tata Steel Ltd.

Abstract

Accurate grade prediction is an important step in the mining planning process. Various methods, namely the Inverse Distance Method and Kriging, are widely used. The application of machine learning is a new development in the grade estimation technique. The present study focused on the application of XGBoost, Random Forests (RFs), Multi Layer Perceptron (MLP), and Gradient Boosting Regression (GBR) models to predict iron ore grades in an Indian mine. An ensemble model was also applied to obtain a more stable grade prediction in the deposit. Models were trained using 4,112 sample data, which have spatial coordinates (east, north, and altitude) and iron grades. The dataset was divided into two parts: 80% (3,289 samples) of the data was used for model training, and 20% (823 samples) was used for model testing. The performance of the models was assessed through the coefficient of determination (R²), Mean Squared Error (MSE), and Mean Absolute Percentage Error (MAPE). The results show that the XGBoost model performs better in the estimation process when compared with other methods, such as RFs, GBR, and MLP. The XGBoost model produced R² of 0.77, MSE of 2.87, and MAPE of 1.8%. The findings indicate that the XGBoost model is effective for predicting iron ore grades in this type of deposit. However, considering geological uncertainty, the application of an ensemble model may be beneficial for grade prediction in an iron deposit.

Keywords

• ML
• Krigging
• XGboost
• Random Forests
• Estimation

Main Subjects

• Surface Mining