Abstract: Machine learning based approaches for Sparse Code Multiple Access (SCMA) have been able to achieve a better performance than the message passing algorithm (MPA) based methods. But these methods have a higher decoding complexity owing to their use of a large number of trainable parameters. This paper proposes a radial basis function (RBF) based SCMA method (RBF-SCMA), which uses an RBF module to efficiently exploit the distance based structure of the decoding problem. This allows it to arrive at better estimates for user symbol probabilities, even with a single hidden layer neural network. Simulations show that RBF-SCMA performs better than the previous methods in an additive white Gaussian noise (AWGN) channel while maintaining a lower computational complexity.