A predictive modeling method used to interpolate multidimensional space, using a hidden layer (often Gaussian) and an output layer, associating each input data point with a Radial Basis Function (RBF). RBM is resistant to local minima problems, but the input space must be covered well by RBFs.