Automated Machine Learning Assigned to Predict Porosity and Permeability: A Case Study, Southern Iraq

Porosity and permeability are considered key inputs in reservoir modelling. Direct measurements of these petrophysical properties are costly and time-consuming. Machine learning is often better than traditional theoretical and statistical techniques in prediction, but selecting the optimal algorithm and hyperparameters can be complicated. This study aims to use the automated machine learning approach to predict porosity and permeability with high accuracy and minimal human intervention. Two sequential models are developed to predict porosity and permeability. These models are selected based on an AutoML framework, known as AutoGluon, which evaluates algorithms like linear, neural networks, and decision tree-based models using conventional log data inputs: neutron, sonic, density, gamma ray, and resistivity logs. The approach was validated on data from three wells in an untapped field in southern Iraq. The AutoGluon approach achieved 86% accuracy for porosity and 80% for permeability. These findings demonstrate that automated machine learning can simplify the workflow by model selection and hyperparameter optimization, which improves prediction accuracy and enhances decision-making on more accurate models.