Abstract
This study proposes a machine learning–based approach for predicting reservoir fluid density using the Extreme Gradient Boosting (XGBoost) algorithm. A dataset comprising 362 experimental measurements from multiple Kazakhstani reservoirs was utilized. Pressure (P), oil formation volume factor (Bo), and solution gas-oil ratio (Rs) were selected as input variables, while fluid density served as the target parameter. After data preprocessing and hyperparameter optimization, the XGBoost model achieved a coefficient of determination R² of 0.80 and a root mean squared error (RMSE) of 0.031, outperforming conventional linear and polynomial regression models. To improve interpretability and engineering applicability, K-Means clustering was applied to identify natural groupings of reservoir fluids, and Principal Component Analysis (PCA) was used for dimensionality reduction and visualization. In addition, a simplified linear correlation was derived using Ordinary Least Squares (OLS) regression based on XGBoost predictions, providing a fast and practical estimation tool for field applications. The proposed methodology combines high predictive accuracy with practical usability and demonstrates strong potential for application in reservoir simulation, production optimization, and the development of regional ML-based PVT models for the oil and gas industry of Kazakhstan.