A Special Supervised Learning Algorithm and Its Applications

The supervised learning algorithm with one normal sample in every class is used in many domains. However the algorithm isnpsilat normal and will influence the result. Because there is only one normal sample in every class, the learning stage is omitted. Very strong information conditions make this recognition algorithm different from other supervised learning algorithms. In this paper we point out: (1)The usual method using dasiacertain distancepsila as similarity measure between samples and normal class samples is not suitable for present information conditions. (2) The best choice is to use membership as similarity measure between index value and classes. However, topological space structure, corresponding algebraic properties and normalization method of the membership must be normalized. (3) The conversion of index membership to sample membership is a pure mathematical problem. The index weight is determined by the information entropy of index value membership and has nothing to do with the decider. A new supervised learning algorithm is proposed and water quality evaluation is solved by this method. The evaluation results using different evaluation methods show the new method is effective.