CO2 dissolution in water at different temperature and pressure conditions is of essential interest for various environmental, geochemical, and thermodynamic related problems. The topic is of special interest in studies of CO2 geological sequestration in brine-bearing aquifers. In this Article, four powerful machine learning (ML) techniques-Radial Basis Function Neural Network (RBFNN), Multilayer Perceptron (MLP), Least-Squares Support Vector Machine (LSSVM), and Gene Expression Programming (GEP)-are implemented to develop economical, rapid, and reliable models to predict the solubility of CO2 in water. To expand the prediction capability of the ML approaches, their control parameters are optimized by various techniques. To this end, four back-propagation algorithms are applied in the MLP learning phase, while Particle Swarm Optimization (PSO), Differential Evolution (DE), Genetic Algorithm (GA), and Firefly Algorithm (FFA) are used to optimize the RBFNN and LSSVM control parameters. A wide-ranged database including temperature and pressure as inputs and CO2 solubility in pure water as output is utilized to develop the models, which are then compared with each other and also against existing models. The results demonstrate that the prediction performance of the proposed models is quite satisfactory. In addition, the comparison results reveal that the LSSVM-FFA is the best paradigm to estimate the solubility of CO2 in pure water, as it outperforms the other proposed ML techniques as well as prior models. The overall RMSE and R-2 values for LSSVM-FFA are 0.3261 and 0.9930, respectively.