A model is implemented that includes one or more classes. For each of the one or more classes of the model, a set of input variables of the big data set are represented as a matrix with non-zero values only provided as diagonal entries. A most likely class for each input variable is determined based at least in part on inverting the matrix. One or more predictions are determined for one or more output variables based at least in part on the most likely class of one or more input variables from the set of input variables.