Knowledge from data ? a new tool for analysis of diverse data

Abstract

Self Organizing Maps (SOM) are a tool for the visualization and interpretation of complex, multivariate data with many potential applications in geophysics. We demonstrate the ability of the SOM method to analyse a data set with 50% missing data. This robustness of a SOM enables it to also be used to interpolate missing data and see through noise to underlying relationships between the data variables. A study involving geophysical log data for the characterization and prediction of lithologies with anomalous levels of radioactivity in the Bowen Basins is used to illustrate the calibration of a SOM. In this case the SOM was used to identify the radioactive sandstones. This was then used successfully as a classification template for data from another mine site, >30km to northeast.