Viscosity emerges as a physical property of primary importance in the modeling of flow within a porous medium, as well as in the processes of production, transport, and refining of crude oils. The direct measurement of viscosity is carried out through laboratory tests applied to samples extracted from the bed of a well, being these samples characterized by their difficult collection and the considerable time lapse required for their acquisition. Several techniques have been developed to estimate viscosity, among which the empirical correlation with Nuclear Magnetic Resonance logs stands out. This study presents a methodology for creating a representative predictive viscosity model, adapted to specific reservoir conditions, using measurements and well logs using machine learning techniques, in particular, Support Vector Machines (SVM). It is concluded that SVM trained with a polynomial kernel (R² = 0.947, MSE = 631.21, MAE = 15.16) exhibits superior performance compared to SVM trained with linear and RBF kernels. These results suggest that SVMs constitute a robust machine-learning technique for predicting crude viscosity in this context.