Comparative analysis of feature selection-based machine learning techniques in reservoir characterization

Comparative study and analysis of the generalization performance and predictive capability of machine learning (ML) techniques in reservoir characterization and modeling is presented in this work by utilizing two distinct Oil well data sets. The performances of the ML techniques (artificial neural network (ANN) and support vector regression (SVR)) have been boosted by proposing a correlation-based feature selection approach which employs fewer datasets resulting in less computing time and processing power. Predictive accuracy of both ML techniques in permeability prediction has been improved as a result of the feature-selection approach. Furthermore, ANN shows superior performance in case of large dataset while SVR shows better performance in case of small dataset. The results of this work provide excellent insight and guidance to practitioners in improving ML performance and determining the most appropriate technique in cases of small and large datasets.