ABSTRACT

Porosity is one of the most critical parameters in reservoir characterization, as it directly influences hydrocarbon storage capacity. Accurate porosity prediction becomes even more essential in fields with limited core data, such as Field “X”, located in the South Sumatra Basin. This study compares two different porosity prediction approaches: a deterministic method based on well log interpretation using NPHI and RHOB logs, and various Machine Learning (ML) algorithms, including Random Forest (RF), K-Nearest Neighbor (KNN), Gradient Boosting (GBR), AdaBoost (ADA), Support Vector Machine (SVM), and Decision Tree (DT). Data preprocessing involved feature selection using Pearson, Spearman, and Kendall correlation coefficients to identify the most influential log parameters. The dataset was then divided into training (70%) and testing (30%) subsets. Model performance was evaluated using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The deterministic method yielded an MAE of 0.0658 and RMSE of 0.0906, while the best ML model, Random Forest, achieved an MAE of 0.0329 and RMSE of 0.0434 on the testing dataset. In conclusion, Machine Learning, especially the Random Forest model, proves to be a more reliable and accurate tool for porosity prediction in geologically complex fields, offering significant potential for enhancing reservoir modeling and field development planning.