An optimal algorithm for pattern classification†

Publisher Website
Google Scholar
Abstract
This paper presents a new optimality criterion and an optimal algorithm to the problem of pattern classification. It is shown that the optimal pattern classification problem may be treated as a linear programming problem by a suitable transformation. The optimal separating surface obtained is unique and the tolerance is maximum. In the case the maximum tolerance turns out to be zero, the algorithm may yield information concerning the linear separability of the training patterns. Numerical examples and corresponding maximum tolerances are also presented.

This publication has 7 references indexed in Scilit: