Committee of the Combined RBF-SGTM Neural-Like Structures for Prediction Tasks

摘要

The paper describes the committee of non-iterative artificial intelligence tools for solving the regression task. It is based on the use of high-speed neural-like structures with extended inputs. Such an extension involves the combined use of primary inputs and extended inputs, via RBF. The resulting combination of inputs allows increasing the extrapolation properties of each element of the committee. This ensures a decreasing of the prediction errors for the solution of the regression tasks in cases of large volumes of data processing. The developed committee is used to solve the task of prediction of insurance costs. It is experimentally found that the proposed committee decreases training and test errors compared with the use of one neural-like structure of this type. The comparison of the committee's effectiveness with existing iterative and non-iterative computational intelligence methods has confirmed the highest accuracy of its work with a small increase of the time of the training procedure. The developed committee in software or hardware variants can be used to solve regression and classification tasks in the condition of large volumes of data for different application areas.