Data Science for Pandemic Response

In March 2020, a multidisciplinary volunteer team began working with the Presidency of the Valencian Government to support its response to COVID-19. The collaboration brought together researchers from across the Valencian university and health systems, with contributions from public institutions, mobile-network operators, and technology partners. Nuria Oliver coordinated the team in her role as Commissioner for the Valencian AI Strategy and Data Science against COVID-19.

The work developed along four connected strands: mobility analysis, epidemiological modeling, forecasting, and citizen science. This page brings together the program and its main research outputs. The original Valencia IA4COVID team page, in Spanish, records the participating institutions and researchers.

The program's most prominent international recognition was winning the Cognizant-sponsored $500K XPRIZE Pandemic Response Challenge, a four-month global competition to develop data-driven systems that could help decision makers respond to COVID-19.

Population surveys and social effects

The COVID19ImpactSurvey was a 24-question online survey about social contact, health, employment, economic effects, and people's ability and willingness to remain confined. It eventually collected more than 500,000 responses across Spain, Italy, Germany, and Brazil.

The first peer-reviewed study analyzed 156,614 responses gathered during Spain's first wave. Later studies examined changes in willingness to remain confined, social isolation, and the practical operation of testing, tracing, and isolation in Spain and Italy. These were voluntary online surveys rather than probability samples, so their results describe the respondents and associations among variables, not exact national prevalence or causal effects.

Selected publications:

Mobility, citizen data, and privacy

Aggregated mobility data can help describe changes in movement and physical contact, but it is not a census. Phone access, operator coverage, and patterns of use can leave some populations underrepresented. This part of the program examined possible public-health uses while addressing governance, privacy, data access, and the need for citizens to understand and control how their data are used.

ACDC-Tracing explored a specific privacy-preserving design: anonymous vouchers that people testing positive could pass to known contacts. Those contacts could seek testing without a central service collecting their identities or location histories. It was a system proposal, not evidence from a population-scale deployment.

Selected publications:

Outbreak analysis and forecasting

One study analyzed all tracked outbreaks in the Valencian Region between September and December 2020. Most recorded outbreaks were controlled within two weeks, while a small number lasted for more than two months. A later recurrent-neural-network model predicted daily cases across 30 countries, incorporating interventions, vaccination, and waning immunity.

Selected publications:

Winning the $500K XPRIZE Pandemic Response Challenge

A defining achievement of the program: the ValenciaIA4COVID team won the Cognizant-sponsored $500K XPRIZE Pandemic Response Challenge.

The forecasting work fed into a predictor and intervention prescriptor. Rather than produce one supposedly ideal policy, the system generated alternatives representing different trade-offs between expected infections and the social and economic cost of restrictions. The winning system was tested with real data during the competition and through the team's collaboration with the Valencian Government.

The paper describing the predictor and prescriptor also received the Best Applied Data Science Paper Award at ECML-PKDD 2021. The XPRIZE project page preserves the original interactive presentation.

Lessons carried forward

Surveys can reveal experiences absent from administrative records, mobility data can improve situational awareness, and models can help compare possible actions. None of these sources is complete on its own. Their responsible use requires attention to representativeness, privacy, uncertainty, transparency, and the difference between prediction and causation.

A later interview study with experts in Spain and Italy found that non-traditional data helped provide rapid information when conventional sources were delayed. It also identified persistent problems with access, quality, demographic coverage, infrastructure, and governance.