Epistemic Uncertainty and Model Transparency in Rock Facies Classification Using Monte Carlo Dropout Deep Learning

Although Deep Learning (DL) architectures have been used as efficient prediction tools in a variety of domains, they frequently do not care about the uncertainty predictions. This may prevent them from being practical applications. In seismic reservoir characterisation, predicting facies data is typically viewed an inverse quantification issue. The goal current study to analyse dependability rock classification model order quantify while maintaining high accuracy by using and evaluating monte carlo dropout based deep learning (MCDL), computationally technique. proposed method unique since it can epistemic classified blind or unseen well conditioned on Seismic attributes bayesian approximation achieved MCDL framework. findings show that MC successful terms reliability, with test F1-scores 98% 82% synthetic datasets respectively. Moreover, applications 2D section indicate internal regions sections are generally less than their boundaries, calculated different realizations network. For comparison, plain DL support vector machine (SVM) also implemented suggest our outperformns other models comparison which indicates potential be robust classification.