The Analysis of Countries’ Investment Attractiveness Indicators Using Neural Networks Trained on the Adam and WCO Methods

Abstract

The urgent task of using new approaches to analyze the foreign direct investment and macroeconomic indicators that affect the volume of their attraction to a particular country in the world economy was solved by the connectionist-metaheuristic approach. The proposed connectionist-metaheuristic approach makes it possible to improve the quality of the approximation due to: simplification of structural identification through the use of only one hidden layer of neural network models; reducing the computational complexity of parametric identification and ensuring good scalability through the use of batch mode for non-recurrent neural network models and multi-agent metaheuristics for recurrent neural network models; descriptions of nonlinear dependencies through the use of neural network models; high approximation accuracy due to the use of recurrent neural network models; resistance to data incompleteness and data noise due to the use of metaheuristics for parametric identification of recurrent neural network models; lack of requirements for knowledge of distribution, homogeneity, weak correlation, and optimal factors’ choice. In the case of using a GPU, you should choose an LSTM-based neural network that has the highest approximation accuracy. For LSTM, the determination coefficient using the Adam method is 0.785, and using metaheuristics (modified wasp colony optimization) is 0.835. The proposed approach makes it possible to expand the scope of approximation methods’ application based on artificial neural networks and metaheuristics, which is confirmed by its adaptation for an economic problem, and contributes to an increase in the intelligent computer systems efficiency for general and special purposes.