Ensemble Machine Learning for Seismic Attribute-Based Porosity Prediction and Lithofacies Classification in Clastic Hydrocarbon Reservoirs: A Comparative Workflow with Uncertainty Assessment
Accurate prediction of reservoir porosity and reliable lithofacies classification are fundamental to hydrocarbon exploration and reservoir development because they directly influence reserve estimation, well placement, and production optimization. Conventional seismic interpretation methods often struggle to capture the complex nonlinear relationships between seismic attributes and reservoir properties, particularly in heterogeneous clastic formations. This study presents an integrated machine learning workflow for simultaneous porosity prediction and lithofacies classification using post-stack seismic attributes calibrated with well-log observations. Twenty seismic attributes representing amplitude, frequency, phase, geometric, and textural characteristics were extracted from a three-dimensional seismic volume and screened using a systematic feature-selection strategy. Four supervised machine learning algorithms, namely Random Forest (RF), Support Vector Regression (SVR), Gradient Boosting Regression (GBR), and Artificial Neural Networks (ANN), were developed and compared for porosity estimation, while corresponding classification models were evaluated for lithofacies prediction. Model performance was assessed using k-fold cross-validation and blind-well validation to ensure robust generalization. Predictive uncertainty was quantified through ensemble-based confidence estimation and incorporated into the final reservoir property volumes. Illustrative placeholder results indicate that ensemble learning algorithms consistently outperform conventional regression approaches by effectively capturing nonlinear relationships among seismic attributes while providing improved porosity prediction accuracy and more reliable lithofacies discrimination. The proposed workflow integrates feature selection, comparative machine learning evaluation, blind-well validation, and uncertainty assessment into a unified framework that can be readily adapted to other clastic hydrocarbon reservoirs. This study demonstrates the potential of modern machine learning techniques for quantitative seismic reservoir characterization while providing confidence-aware predictions for exploration and field-development decision making.