Hyper-Rectangular and k-Nearest-Neighbor Models in Stochastic Discrimination