Data Mining Mining Data - Ordered Vector Quantisation and Examples of its Application to MIne Geotechnical Data Sets

Computational techniques are needed to assist in the analysis and interpretation of the increasing amounts of geoscientific and mining related data and information that are routinely gathered at mine sites. New interpretation methods are needed to provide an integrated approach so as to establish relationships (cause and effect) or to be predictive about possible outcomes based on acquired data. Traditional multivariate statistical methods are often confused by variable relationships that are non-linear, data distributions that are non-normal, and by the data themselves that may contain missing values, text and both continuous and discontinuous numeric values. Modern data mining methods are available to understand the relationships within and between diverse data sets, and identify relationships or trends associated with processes, measurements and mine design parameters in mines. The self-organizing map (SOM) is a data mining approach with the advantage that all input data samples are represented as vectors in a data-space defined by the observations (variables). The SOM procedure is an exploratory analysis tool that highlights patterns and relationships. Results are internally derived, in an unsupervised fashion based on measures of vector similarity. Our SOM outputs are highly visual, which assists in understanding and illustrating the dataÆs structure and internal relationships. Two studies are presented. The first uses SOM to investigate the relationships between a range of measured and derived rock property parameters and their observed behaviours during mining. Our aim was to determine whether a SOM analysis would allow a better understanding of observed and predicted rock behaviours from their rock properties. The SOM analysis showed that there were indeed relationships between æbrittleÆ, æsugaryÆ, æslabbyÆ and æsqueezingÆ rock behaviours and a number of the input parameters. The computed SOM æstructureÆ was then used to assess the likely behaviour of unknown samples by assigning them into behavioural fields based on their measured and derived geotechnical responses. The second study was aimed at assessing whether a SOM analysis of microseismic data at the Kalgoorlie Consolidated Gold MineÆs (KCGM) Mt Charlotte mine can contribute towards understanding and potentially forecasting/minimising seismicity at the mine-design stage. The database comprised 40 sample æclustersÆ of microseismic parameters from the MS-RAP software package, which were derived from 2500 individual seismic events. The SOM analysis showed that some of the initial mining conditions are related to the seismic parameters. In each of the studies, the SOM procedures assisted in understanding the data and provided new insights. Because of its æordered vector quantisationÆ foundation, the SOM procedure is a substantial improvement over traditional statistical methods in three main areas:its ability to analyse data from disparate sources, including different data types; its ability to analyse sparse data sets with missing values; and its ability to analyse and visualise non-linear relationships within a data set.