In this paper is carried out a comparative study between Self-Organizing Maps (SOM) and Bayesian Networks, to evaluate their performance in the field of structural health as damage detector of structural failures type III, which detects where the damaged area is and percentage of damage within this the area. The implemented classifiers are trained to detect structural failures in a simply supported beam and a truss of 13 elements, the detection is performed using modal information from different test scenarios, obtained by OpenSees® and MATLAB®. The simulations show a satisfactory performance of Bayesian networks to provide the range of stiffness variation in each element of the studied systems. Meanwhile, SOM networks are useful in predicting the decrease in elastic modulus, which is assumed as specific percentage of damage. Based on this results is proposed a hybrid methodology (BAYSOM) seeking to reduce computational cost and improve performance in diagnosis and detection of fault conditions in structures.