On Approximation of Binary Random Vectors in Discriminant-Analysis Problems

With reference to the construction of a discriminant function, this paper makes a comparison of the approximation quality of the probability distributions of binary random vectors for which the number of independent parameters is $2^l-1$. A comparison is made of the Bahadur function and the probability density function of the normal distribution law, with number of parameters equal to $l(l+1)/2$. It is shown that these approximations are equally effective for setting up a classification rule; this makes it easier to investigate binary random vectors and random vectors containing binary and continuous features.