Deep learning approaches to mid-term forecasting of social-economic and demographic effects of a pandemic

Dmitry Devyatkin, Yulia Otmakhova, Natalia Usenko, Ilya Sochenkov, Vladimir Budzko

Research output: Contribution to journalConference articlepeer-review


The COVID-19 outburst has brought serious demographical, economic, and social impacts. Moreover, in large countries, these consequences can vary from region to region. Therefore, authorities and experts lack the models to predict these various impacts at the regional level. This paper presents deep neural network models to do a mid-term forecast of the COVID-19 effect in the Russian regions. The models are based on the various recurrent and sliding-window architectures and utilize the attention mechanism to consider the indicators of the neighbor regions. These models are trained on various data, including daily cases and deaths, the diseased age structure, transport availability of the regions, and the unemployment rate. The experimental evaluation of the models shows that the demographic and healthcare indicators can significantly improve mid-term economic impact prediction accuracy. We also revealed that the neighboring regions' data helps predict the pandemic's healthcare and demographical impact. Namely, we have detected improvement for both the number of infected and the death rate.

Original languageEnglish
Pages (from-to)156-163
Number of pages8
JournalProcedia Computer Science
Publication statusPublished - Jul 2021
Event2020 Annual International Conference on Brain-Inspired Cognitive Architectures for Artificial Intelligence: Eleventh Annual Meeting of the BICA Society, BICA*AI 2020 - Natal, Rio Grande do Norte, Brazil
Duration: 10 Nov 202015 Nov 2020


  • attention mechanism
  • COVID-19 pandemic
  • Long-Short Term Memory
  • mid-term impact prediction
  • recurrent neural network
  • socio-economic impact




Dive into the research topics of 'Deep learning approaches to mid-term forecasting of social-economic and demographic effects of a pandemic'. Together they form a unique fingerprint.

Cite this