Methodology of Building a Neural Network Architecture for Regulating Social and Labour Processes in BRICS Megacities

Abstract

The article presents a comprehensive methodology for building a neural network architecture for forecasting, adapting and regulating social and labour processes in the BRICS megacities, where approximately 470–480 million people lived in 2023. The study substantiates the use of hybrid architectures based on RNN, LSTM, CNN and GAN, which ensures the processing of temporal, spatial and synthetic data in highly urbanized environments. The level of urbanization in the BRICS countries ranges from 36% in India to 87% in Brazil, which requires customized digital solutions. It has been revealed that informal employment in India reaches 80%, and in South Africa the unemployment rate in 2023 was 32%, which creates the need for models to restore hidden labour indicators. The authors demonstrate that the use of attention mechanisms allows taking into account country-specific features, and explicable AI increases the transparency of decisions for government authorities. Special attention is paid to platform employment: up to 46% of workers in Brazil and India face unstable orders, while in Russia and China it is about 31%. Generative networks (GANs) are used to model social policy scenarios taking into account stress factors. Special attention is paid to the Decent-Gig Index metric as a target indicator for training samples. The institutional asynchrony between the BRICS countries is offset by the multitasking architecture that supports different legal regimes. It is shown that neural networks can interpret migration flows, for example, 140 million seasonal workers in India annually. The article highlights the importance of digital ecosystems of megacities in shaping flexible employment policies. The proposed architecture is focused on integration into urban management systems through APIs and multi-agent platforms. The possibility of using real-time labour analytics, already implemented in Shenzhen and Sao Paulo, is substantiated. The methodology is based on 18 sources with up-to-date data and confirms the high scientific and practical importance of using neural networks in regulating social and labour relations.