Autor: |
Klym Yamkovyi, Galyna Grinberg, Leonid Lyubchyk |
Rok vydání: |
2018 |
Předmět: |
|
Zdroj: |
2018 IEEE First International Conference on System Analysis & Intelligent Computing (SAIC). |
DOI: |
10.1109/saic.2018.8516730 |
Popis: |
The problem of integral indicator for complex system building based on aggregation of a set of partial indexes is considered. Within semi-supervised learning concept, it is assumed that training dataset consists of a group of objects with measured values of partial indexes and expert estimates of the corresponding values of the integral indicator and a group of objects, for which expert information is not available. For estimation of linear model parameters the method of optimal concordation of both partial indexes relative significance weights and integral indicator values is used. Unlabeled dataset provides additional regularization using the graph data model by smoothing of desired indicators model on data cloud. A nonlinear model is built on the basis of kernel-based model approach with regularization by optimal concordation with linear model parameters estimates. An unlabeled dataset is used for kernel functions transformation for model smoothing considering data geometric structure. |
Databáze: |
OpenAIRE |
Externí odkaz: |
|