Robust polynomial neural networks in quantative-structure activity relationship studies

Authors:
Tetyana I. Aksyonova;Vladimir V. Volkovich;Igor V. Tetko
Affiliations:
Laboratory of Applied Nonlinear Analysis, Institute of Applied System Analysis, prosp Peremogy, 37, 03056, Kyiv, Ukraine;Control System Department International Researching-Training Center of Information Technologies, Glushkova 40, 252022, Kyiv, Ukraine;Laboratoire de Neuro-heuristique, Institut de Physiologie, Université de Lausanne, Rue du Bugnon 7, CH-1005, Lausanne, Switzerland
Venue:
Systems Analysis Modelling Simulation - Special issue: Self-organising modelling and simulation
Year:
2003

Citing 2
Cited 5

Neural Networks in Chemistry and Drug Design

Neural Networks in Chemistry and Drug Design
Inductive Learning Algorithms for Complex Systems Modeling

Inductive Learning Algorithms for Complex Systems Modeling

Self-organising modelling as a part of simulation process

Systems Analysis Modelling Simulation - Special issue: Self-organising modelling and simulation
Direct transformation of coordinates for GPS positioning using the techniques of genetic programming and symbolic regression

Engineering Applications of Artificial Intelligence
Robust structural modeling and outlier detection with GMDH-type polynomial neural networks

ICANN'05 Proceedings of the 15th international conference on Artificial neural networks: formal models and their applications - Volume Part II
Discovering approximate expressions of GPS geometric dilution of precision using genetic programming

Advances in Engineering Software
Survey: Multi-criterion Pareto based particle swarm optimized polynomial neural network for classification: A review and state-of-the-art

Computer Science Review

Quantified Score

Hi-index	0.00

Visualization

Abstract

This article presents the Robust Polynomial Neural Networks, a self-organizing multilayered iterative GMDH-type algorithm that provides robust linear and nonlinear polynomial regression models. The accuracy of the algorithm is compared to traditional GMDH and the multiple linear regression analysis using artificial and real data sets in quantitative-structure activity relationship studies. The calculated data shows that the proposed method is able to select nonlinear models characterized by a high prediction ability, it is insensible to outliers and irrelevant variables and thus it provides a considerable interest in quantitative-structure activity relationship studies.