<p>Accurate prediction of hydrocarbon solubility in ionic liquids (ILs) is essential for optimizing solvent selection and designing efficient chemical processes in industries such as energy production and environmental remediation. ILs offer unique advantages, including low volatility, tunable structures, and high solvation capacity, but their complex interactions with solutes make solubility prediction challenging. This study employs Gaussian Process Regression (GPR), a probabilistic machine learning method that enables accurate predictions and uncertainty quantification, enhanced with four kernel functions: Matern, exponential, squared exponential, and rational quadratic. A dataset of 1,018 experimental solubility measurements was compiled from the literature to train and validate the models. The GPR model with the Matern kernel achieved outstanding predictive performance with an R² value of 0.995, while the exponential, squared exponential, and rational quadratic kernels also demonstrated high accuracy, with R² values of 0.994, 0.992, and 0.992, respectively. Sensitivity analysis identified pressure with 63% relevancy factor as the most influential factor affecting solubility. Furthermore, the GPR models outperformed the widely used COSMO-RS thermodynamic model, demonstrating superior accuracy and computational efficiency. Our findings suggest that AI-driven approaches offer a clear advantage over conventional thermodynamic models for predicting gaseous hydrocarbon solubility in ionic liquids.</p>