This study introduces a machine-learning-driven workflow to optimize CO2-WAG by considering both oil production and CO2 sequestration as the output parameters. A compositional simulator is employed to create a baseline model by using data from the Bell Creek Oilfield. Computational models, specifically multi-layer perceptron (MLP) combined with evolutionary and metaheuristic algorithms, such as the genetic algorithm (GA), particle swarm optimization (PSO), and reptile search algorithm (RSA), are utilized to develop proxy models for predicting cumulative oil production and CO2 storage. The suggested workflow integrates the optimization algorithms GA, PSO, and RSA with MLP-based proxy models (MLP-GA, MLP-PSO, and MLP-RSA) to enhance cumulative oil production and CO2 storage. The findings demonstrated the reliability and efficient utilization of the proposed workflow for CO2-WAG coupled optimization. All proposed models surpassed the base model; among them, MLP-RSA outperformed MLP-PSO and MLP-GA in terms of prediction accuracy and optimization performance. In comparison to the base case, the optimal scenarios increased the cumulative oil production by 10.65% and CO2 storage by 24.72%. Among the optimization algorithms, RSA outperformed the others in terms of speed, processing, and yielded an incremental oil revenue of 15.31 million USD. The results of this study offered new insights into optimization techniques and decision-making processes for CO2-WAG projects. Moreover, it provides engineers with a tool to conduct robust and effective optimization studies, avoiding tedious conventional optimization techniques.