Abstract A new mathematical model was proposed integrating the algorithms of neural networks and gradient boosting for the selection of the optimal EOR method in heavy and extra heavy oil fields as well as to predict the oil production of the field in the first five years of the project. The parameters were selected from genetic algorithms using the technique of auto-machine learning. The model was developed with a database of 919 fields, from different public sources, and consists of three stages: 1) A dimensional reduction of the initial variables using t-SNE (t-Distributed Stochastic Neighbor Embedding), passing from 12 variables to 2 final variables. 2) An estimation of the optimal EOR method from an Artificial Neural Network. And 3) the integration of 1) and 2), serve as the input for a genetic algorithm in order to forecast the production of the first years of the field after the project started. The obtained results show a correlation of 87% for the classification neural network, and 93% for the genetic algorithm in the test database. Therefore, an integrated workflow was also developed, which takes the generated production profile and performs cash flow analysis, leading to the calculation of key financial parameters for determining project feasibility from given reservoir, fluid, and economic conditions input by the user. The developed model is presented as an opportunity to experience the possibilities that artificial intelligence has in the oil and gas industry, and how it can be of great help in the prediction of variables for process optimization. Employing the optimal EOR method from the beginning of field development can increase long-term profitability. In addition, the developed screening software eases decision making for operators in selecting, technically and economically, the most feasible thermal EOR method for the development of heavy and extra heavy oil fields, maximizing asset value and E&P portfolio.