​Micromine 2016 Feature Highlight-Improved Cross Validation tool now includes IDW​

anonymous (Moderator (EN)) 8 years ago in General updated by anonymous 7 years ago 0

Cross validation has been available under Stats | Cross Validation for some time but it has been overhauled for Micromine 2016. New and improved features in Cross Validation allow you to:

  • Validate Inverse distance weighting (IDW) procedures as well as Kriging.
  • Justify all of your estimation choices including interpolation method, semivariogram model, and search neighbourhood and strategy.
  • Automatically display the cross validation results in Vizex for quick and easy interrogation and interpretation.

Cross validation is used to test modelling parameters prior to estimation. It is completed by removing a point from the data, estimating its value using the rest of the data and the chosen interpolation method, model or search neighbourhood, and then comparing this estimate to the removed data point. This is completed sequentially for each point in the data set.

The statistics for each data point are output in a data file and the cumulative statistics are output in a report file.

If you are using Kriging the report file will include the mean and standard deviation of the raw data, the standard error, the estimate, the residual and the error statistic, and also the Root Mean Square Error (RMSE).

New to Micromine 2016 is cross validation of Inverse Distance Weighting (IDW) methods. If you choose this method the report file will contain the mean and standard deviation of the raw data, the estimate, as well as the RMSE.

Image 1002

Cross validation dialog now with IDW and option to Auto load results into Vizex with sizing and colour coding

For Kriging, the mean and standard deviation of the error statistic, and for IDW the RMSE, are used to validate the estimation parameters.

However, it is not sufficient to rely solely on these statistics. Using the output data file and the Micromine 2016 Stats menu you can complete a range of further analyses such as summary statistics and histograms of the estimated values, scattergrams and QQ-plots of the actual and estimated grades, histograms of the error statistic and residuals, and scattergrams of the error statistic versus the actual grades.

A vital step is to view the error statistic in 3D to check for spatial bias. Visualisation of the error statistic is much improved in Micromine 2016 as there is now an option to Auto load the data points scaled and colour coded by the error statistic (Kriging) or Residual (IDW). This allows you to instantly view the results of the cross validation in context with your drilling data and interactively query them file using Sync Selection.

Image 1003

Cross validation results of coal seam thickness modelling displayed as points colour coded and sized by error statistic.