Water alternating carbon dioxide (WAG) synergy of enhanced oil recovery (EOR) has emerged as the most environmentally safe method to increase the recovery of oil from petroleum reservoirs. Despite reported incremental oil recoveries of WAG, not much attention has been dedicated to the Low Salinity Water-Alternating-Carbon Dioxide (CO2 LSWAG) form of WAG for EOR in carbonate reservoirs. Also, neither the quality and composition of the water used in most WAG EOR processes are considered, nor are the fast computational methods for future recovery factor predictions. In this paper, multiphase multicomponent flow equations and geochemical modeling (aqueous, mineral, and ion exchange reactions) in a compositional simulator were coupled to generate low salinity ranges with CO2 injection schemes to build proxy models for oil recovery factor predictions under miscible conditions. Additionally, this paper aims at developing workflows of mathematical correlations for CO2 LSWAG oil recovery factor predictions on a field scale using Multivariate Adaptive Regression Splines (MARS) and Group Method of Data Handling (GMDH) supervised machine learning models. Also, to address the uncertainty in CO2 LSWAG test predictions, varied low salinities of 573.86 ppm, 1250.51 ppm, and 2949.15 ppm were used as test data. The GMDH model predicted the oil recovery factor testing data as the best proxy model based on statistical measures employed compared to the MARS model. The CO2 LSWAG proxy models developed in this study optimize and reduce simulation run times compared to the traditional numerical simulation without compromising the obtained accuracies. This study strongly recommends the injection of CO2 LSWAG in carbonate reservoirs due to its high recovery factor for improving microscopic and macroscopic displacement efficiencies. © 2024 Daniel Asante Otchere.