Indexing of objects on the individual sets of informing features

Criteria are investigated for the estimation (indexing) of object on the individual set of different-type features. It is considered, that a training sets of E₀={Si,... ,S_m} divided into of disjoint subset (classes) K_bK₂. The objects of selection are described by a set of different-type features ofX(n), distances between objects are calculated on the metric of Juravlev. The individual set of informing features of possible object allows: - to distinguish logical regularities in his neighborhood and calculate their stability; - to explain a decision-making process at recognition; - to determine belonging to the noise objects of classes; - to produce the choice of supporting sets of features in the models of algorithms of recognition. For the search of informing set of features of X(k) с X(n), k 0. The value of estimation of object of SeE₀ is calculated as extremum of criterion on the individual set of X(k). A heuristic allowing to decrease combinatory complication of algorithm of extraction of features on an additive criterion offers. The method of synthesis of own space is worked out from latent features for description of object of SeE₀ on the basis of hierarchical clustering. Every latent feature is result of nonlinear mapping of group of initial features on the real axis. The sequence of forming of latent feature the algorithm of hierarchical clustering is brought around to an example: x₂ = 0,1 (1,6813 (x2 + 0,0175)-0,2426xJ₀) +1,7133 (x2x₁0 + 0,0175); x3 = 1,0118 (x2 + 0,0242)-0,2144x₇. The estimations of objects can be used for the construction of models in weakly formalizing subject domains. A requirement in using on the individual set of informing features of object for making decision arises up at raising of diagnosis of illness, to development of measures on prevention of technogenic catastrophes on concrete territory. Values of estimations in [0,1] are easily interpreted in terms of fuzzy logic and can be used for verbalization of knowledge.

Keywords

Authors

References