Enhanced Machine Learning Modelling Techniques for Better Classification of Carbonate Reservoir Rocks

Abstract This study aims to enhance machine learning models for classifying carbonate rocks into limestone and dolomite using well logging and core analysis data. The research evaluates various machine learning algorithms' performance and identifies effective techniques to improve model accuracy for geological and environmental applications. The study employed several strategies to improve classification models, including grid search, random search, Bayesian optimization, SMOTE, and ensemble techniques (boosting and bagging). A dataset of 4290 points was used to train eight different classification models: Random Forest (RF), Multilayer Perceptron (MLP), Naive Bayes (NB), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Radial Basis Function (RBF), Regression Tree (RT), and Parallel Additive Regression Tree (PART). All implemented strategies significantly enhanced the machine learning classification models, achieving a correlation coefficient near one and a mean absolute error of 0.16 averaged across all techniques. Random Forest and Multilayer Perceptron demonstrated exceptional performance, with accuracy rates of 99.2% and 98.7%, respectively. The Kappa statistic further confirmed the superiority of these models. The study highlights the importance of selecting appropriate machine learning models and optimizing their hyperparameters for effective carbonate rock type classification. The findings underscore the potential for improved accuracy through ensemble methods and hyperparameter optimization in geological classification tasks. This research provides new insights into applying machine learning techniques for geological classification, particularly in carbonate rock type identification. The results have significant implications for developing more accurate and reliable classification models in geoscience applications, potentially improving various geological and environmental studies.