Perpustakaan Digital ITB

This study explores the application of machine learning (ML) algorithms to predict fluid types in deltaic reservoirs, specifically focusing on the Mahakam Delta in East Kalimantan. Deltaic reservoirs are characterized by high heterogeneity and complex geological formations, which pose significant challenges for traditional fluid identification methods. By utilizing well logging data, including parameters such as gamma ray, resistivity, density, and sonic logs, the study applies supervised learning techniques—specifically XGBoost and LightGBM models—to accurately classify fluid types across various wells in the reservoir. The data was pre-processed using imputation for missing values and Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance, ensuring a balanced and robust dataset for model training. The models were evaluated based on metrics including accuracy, precision, recall, specificity, and F1 score, with XGBoost achieving superior performance, showing an accuracy of 91.71% during training and 68.5% during blind testing. The study demonstrates that machine learning techniques can significantly improve the time efficiency and accuracy of fluid type prediction compared to conventional methods, offering a more scalable and reliable approach to reservoir management in complex environments. This work highlights the potential of machine learning in enhancing fluid identification, optimizing exploration efforts, and contributing to the development of more effective reservoir management strategies in deltaic reservoirs.