Classical and Quantum Machine Learning for Population-Level Prediction of Heat-Related Physiological Events
Saúl González-Bermejo, Tommaso Albrigi, Borja Vázquez-Morado, Urko Regueiro-Ramos, Daniel Casado-Faulí, Sergi Cònsul-Pacareu, Parfait Atchadé-Adelomou(1) Lighthouse Disruptive Innovation Group SL · 2026
Read Classical and Quantum Machine Learning for Population-Level Prediction of Heat-Related Physiological Events by Saúl González-Bermejo, Tommaso Albrigi, Borja Vázquez-Morado, Urko Regueiro-Ramos, Daniel Casado-Faulí, Sergi Cònsul-Pacareu, Parfait Atchadé-Adelomou(1) Lighthouse Disruptive Innovation Group SL (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-15382.txt.
Introduction
Gonzalez-Bermejo and six colleagues compare classical and quantum machine learning for predicting heat-related health events across whole populations. Merging climate, demographic, and social data into weekly county-level sets for the United States and Catalonia, they train a classical baseline against a variational quantum model. Classical models still win on accuracy, especially where targets are sparse and skewed, but quantum models show real learning ability.
Catalog record
| CREATOR | Saúl González-Bermejo, Tommaso Albrigi, Borja Vázquez-Morado, Urko Regueiro-Ramos, Daniel Casado-Faulí, Sergi Cònsul-Pacareu, Parfait Atchadé-Adelomou(1) Lighthouse Disruptive Innovation Group SL |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33872 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.15382
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Classical and Quantum Machine Learning for Population-Level Prediction of Heat-Related Physiological Events Saúl González-Bermejo, Tommaso Albrigi, Borja Vázquez-Morado, Urko Regueiro-Ramos, Daniel Casado-Faulí, Sergi Cònsul-Pacareu, Parfait Atchadé-Adelomou(1) Lighthouse Disruptive Innovation Group SL Abstract Predicting heat-related physiological events at the population level is challenging due to the complex interactions among climatic, demographic, and socioeconomic factors, as well as the strong sparsity and seasonality of observational data. In this work, we propose a unified predictive framework that integrates heterogeneous environmental and public-health datasets and evaluates two learning paradigms within a common pipeline: classical machine learning and quantum machine learning. The methodology combines data harmonization, temporal aggregation, feature engineering, and dimensionality reduction to construct a weekly county-level population dataset. On this unified representation, we train both a classical regression baseline and a variational quantum model based on parameterized quantum circuits with angle embedding and data re-uploading. Experimental evaluation on datasets from the United States and Catalonia shows that classical models currently achieve higher predictive accuracy, particularly under conditions of strong class imbalance and sparse targets. Nevertheless, the quantum models demonstrate non-trivial learning capability and capture meaningful predictive structure in several scenarios. These results provide an empirical comparison between classical and quantum learning approaches for population-level physiological prediction and establish a methodological foundation for future hybrid health modeling as quantum hardware continues to evolve. I Introduction Heat-related physiological events and heat-related illnesses constitute an increasingly important public-health concern, particularly under conditions of rising temperatures, demographic aging, urban heat exposure, and heterogeneous social vulnerability [ 8 , 22 ] . Predicting these events at the population level is relevant for public-health preparedness, territorial risk assessment, and resource planning. However, this task remains challenging because heat-related outcomes are driven by the joint effect of environmental conditions, population structure, and socioeconomic context, while the observable data are often sparse, aggregated at mismatched geographic scales, and strongly seasonal. In practice, the prediction of heat-related physiological events cannot be reduced to a simple univariate climatic forecasting problem. Instead, it requires the integration of multiple heterogeneous data sources, including meteorological variables, demographic composition, labor-sector structure, and hospitalization-related records. Moreover, the target variable is typically characterized by strong zero inflation, low absolute counts, and temporal concentration during summer periods. These properties make the learning problem statistically difficult and particularly sensitive to modeling assumptions. Classical machine learning methods offer a natural baseline for this type of task, as they can flexibly model nonlinear relationships from structured tabular data. Boosting-based methods are particularly strong on tabular data and have become standard competitive baselines in many applied prediction settings [ 14 ] . At the same time, recent advances in variational quantum circuits and hybrid quantum–classical optimization have motivated the exploration of quantum machine learning for regression problems, especially in settings where nonlinear feature interactions and compressed latent representations are relevant [ 2 , 20 , 18 ] . Nevertheless, the practical applicability of quantum machine learning in real public-health prediction tasks remains largely underexplored. The present work is therefore framed as a comparative modeling study . The goal is not to claim a quantum advantage, nor to cast the problem as inherently quantum, but rather to establish a common predictive pipeline in which both classical and quantum models can be trained and evaluated under the same data conditions. This allows us to answer a more grounded research question: to what extent can current quantum machine learning architectures approximate or complement classical predictive performance in population-level modeling of heat-related physiological events? To address this question, we construct a unified weekly county-level population dataset from heterogeneous USA and Catalonia data sources. We then apply a shared preprocessing and dimensionality-reduction pipeline and train two model families: a classical regression baseline and a variational quantum model based on a Quantum Sequential Model (QSM). Their performance is compared across multiple population-level prediction settings. The main contributions of this work are as follows: • We propose a reproducible population-level data construction pipeline integrating climatological, demographic, economic, and hospitalization-related information from heterogeneous public sources. • We formalize a common preprocessing strategy that supports both classical and quantum learning, including standardization, correlation analysis, and principal-component compression. • We implement and evaluate both classical and quantum regression models under identical experimental conditions. • We provide an empirical comparative analysis that clarifies where classical methods currently dominate and where quantum models already display meaningful learning behavior under NISQ constraints. II Related Work II-A Population-Level Prediction of Heat-Related Health Outcomes The prediction of heat-related illness, heat-related hospitalization, and heat-sensitive physiological events has traditionally been studied within environmental epidemiology and public-health risk modeling. A substantial portion of the literature focuses on estimating the effect of temperature extremes, heat waves, and humidity-related stress on morbidity and mortality outcomes through time-series regression and distributed lag non-linear models (DLNMs) [ 9 , 3 , 8 ] . These approaches have been instrumental in establishing the association between heat exposure and adverse health outcomes, including hospitalization burdens [ 17 , 15 ] . Although this literature has produced strong explanatory evidence, many studies are not designed as granular predictive systems that integrate heterogeneous environmental, demographic, and socioeconomic data streams. Recent work has started to explore finer-scale geographic variability in heat-related hospitalizations and local vulnerability [ 11 ] , but the predictive integration problem remains open, especially when health outcomes are observed at coarser administrative scales than climate variables. A second limitation in the existing literature is that many studies focus predominantly on environmental exposure variables while incorporating demographic and socioeconomic factors only in a limited or aggregated way. However, vulnerability to heat is known to depend not only on temperature itself, but also on age structure, labor conditions, urban form, and social inequality [ 8 , 6 ] . Therefore, realistic population-level predictive frameworks should integrate climate, demography, and socioeconomic structure jointly rather than treating them as separate analytical layers. II-B Classical Machine Learning for Health and Environmental Prediction Classical machine learning has become increasingly relevant for structured health and environmental prediction tasks, particularly when complex nonlinear relationships are expected among predictors. Methods such as random forests, gradient boosting, support vector regression, and neural networks have been used in environmental health, hospitalization forecasting, and tabular biomedical prediction problems. Among these methods, boosting-based approaches are particularly effective in structured tabular datasets with mixed feature types and nonlinear dependencies [ 14 ] . Their strong empirical performance in applied health prediction makes them a natural benchmark for the present work. However, classical models also face limitations in strongly imbalanced and zero-inflated settings. When the target contains a very high proportion of zeros, standard regression losses may favor conservative predictions, reducing sensitivity to rare but important events. In addition, strong seasonal concentration and low absolute event counts can reduce the effective signal-to-noise ratio, especially after geographic redistribution or aggregation. These issues are directly relevant in the present study. II-C Quantum Machine Learning for Supervised Regression Quantum machine learning (QML) has emerged as a promising line of research at the intersection of quantum computing and statistical learning. Within the current NISQ paradigm, the most widely studied approaches are variational quantum circuits, which are parameterized quantum models trained through hybrid classical–quantum optimization loops [ 2 , 20 , 1 ] . For supervised regression, a