Multiple machine learning algorithms for lithofacies prediction in the deltaic depositional system of the lower Goru...

Machine learning techniques for lithology prediction using wireline logs have gained prominence in petroleum reservoir characterization due to the cost and time constraints of traditional methods such as core sampling and manual log interpretation. This study evaluates and compares several machine learning algorithms, including Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Artificial Neural Network (ANN), K-Nearest Neighbor (KNN), and Logistic Regression (LR), for their effectiveness in predicting lithofacies using wireline logs within the Basal Sand of the Lower Goru Formation, Lower Indus Basin, Pakistan. The Basal Sand of Lower Goru Formation contains four typical lithologies: sandstone, shaly sandstone, sandy shale and shale. Wireline logs from six wells were analyzed, including gamma-ray, density, sonic, neutron porosity, and resistivity logs. Conventional methods, such as gamma-ray log interpretation and rock physics modeling, were employed to establish baseline lithological profiles, while core sample reports provided the necessary ground-truthing of the machine learning models. These traditional interpretations served as a benchmark for evaluating the performance of machine learning algorithms. The results revealed that Random Forest and Decision Tree models outperformed other algorithms, achieving accuracy, precision, recall, and F1 scores in the 96-98% range. Their robustness to noise, interpretability, and ability to handle complex, nonlinear data made them particularly suitable for the heterogeneous and multivariate nature of subsurface data. While SVM, ANN, KNN and LR required more tuning and were prone to overfitting, RF and DT proved efficient and reliable. The integration of traditional geological methods with machine learning provided a comprehensive approach to lithology prediction, enhancing reservoir characterization accuracy, while this study underscores the importance of combining domain expertise with computational models for optimizing petroleum exploration in complex geological environments.