Combining Machine Learning and Scratch Test Data for Enhanced Rock Strength Prediction

This presentation shows the results of applying machine learning methods to scratch test data for the development of robust rock strength models. The core data were modelled using simple regression methods and several machine learning methods (domain transfer analysis, neural network and multi-linear regression). Data of one formation were used for building / training of the models, while the data of the other formation were used for blind testing of the models. The models were subsequently applied to a third formation for rock strength prediction. The machine learning process was subsequently automated using Experienced Eye.

The combination of machine learning methods and scratch test data appeared powerful. Machine learning methods appeared to perform best when more variation in the training data was present. Training the models with scratch test data from both rather than just one formation led to very good results. One single machine learning model could be built from the scratch test data for both sandstone and shale formations. For the given data set, domain transfer analysis appeared to be a more robust method than neural network and multi-linear regression.