Workover rigs are a crucial resource in petroleum exploration and production, used in the wells maintenance operations. The Workover Rig Scheduling Problem (WRSP) determines which rigs will serve the wells and when the activities will occur. This decision-making problem emerges in a highly uncertain environment, and most literature approaches are based on deterministic models and heuristics. Aiming to assist the WRSP, this thesis proposes a regression-based data-driven (DD) optimization methodology, applying it in real-life-based instances. This DD optimization approach is composed of three phases: data treatment, where text mining and clustering techniques are used to refine and retrieve information from the data; predictive modeling, using ridge regression to estimate the workover duration and the endogenous uncertainties in the model; optimization, where the regression prediction and random error are inserted in the joint chance-constrained (JCC) models, generating solutions more resilient to the uncertainties. We propose a stochastic JCC formulation based on simulation and Wasserstein distance to generate scenarios and reduce the problem size. This model is compared with four alternatives: a non-stochastic DD, a stochastic integrated CC, a stochastic budget-constrained model, and the company s current approach. For small and medium-sized instances, the stochastic JCC model guarantees a feasibility confidence level with an error of approximating lower than 5 percent. However, the stochastic JCC model does not close the GAP in large instances. For these instances, the non-stochastic DD model is a good alternative with disturbances not greater than 10 percent. Overall, the DD optimization methodology finds schedules that are more often feasible and with lower costs compared with the company s method.