Applying Data Mining and Mathematical Morphology to Borehole Data Coming from Exploration and Mining Industry

Abstract

Mining companies investigate very carefully the area of proposed mine sites. This is done by first looking at the geology of the area and then drilling the boreholes to predict the quantity and if possible approximate the structure of the mine and distribution of the metal grades. The data obtained from boreholes is analysed using point interpolation techniques such as inverse distance weighting (IDW) or Kriging. However, these techniques have some shortcomings as they heavily rely on strong spatial correlation and they assume linear dependency. In this paper we show how data mining techniques can contribute to planning and even to interpolation tasks when used on borehole data. For this, we first transform the borehole data into a form that is suitable for our methods, then perform k-near-neighbours (k-NN) classification and association rules mining analysis. We also compare k-NN classification method with IDW and show how association rules discovered during the process can improve the results of each method. Moreover, we propose using mathematical morphology operations to filter results for better understanding and, perhaps, for better accuracy. Overall this paper shows possible application of data mining techniques in the mining industry and presents a general framework for carrying out such tasks.