Bloody Forecast: Daily Blood Demand Prediction Using Various Modeling Approaches

Bloody Forecast: Daily Blood Demand Prediction Using Various Modeling Approaches

Stať ve sborníku v databázi WoS či Scopus

Sufficient blood supply is critical not only for scheduled surgeries, but also for emergency medical interventions. In our study, we focus on predicting the daily blood demand separately for two blood types: A+ and O-, based on data from the Transfusion and Tissue Department of University Hospital Brno. The dataset consisted of data on blood demand from 2021 to 2024 and was extended by data regarding non-working days, national and school holidays, seasons, and influenza epidemics. The performance of various prediction models was measured using the normalized Mean Absolute Error (nMAE), which reflects the average prediction error relative to the average daily blood demand. When tested on data from 2023, the best performance was achieved by linear regression models, with a nMAE of 26% for A+ and 50% for O-, indicating lower predictability for blood types with smaller populations. Interestingly, models for different blood types use different features, as the demand for individual blood types depends on different factors. Despite relatively high nMAE values, the models still outperformed a”qualified guess” approach based only on historical averages.

Sufficient blood supply is critical not only for scheduled surgeries, but also for emergency medical interventions. In our study, we focus on predicting the daily blood demand separately for two blood types: A+ and O-, based on data from the Transfusion and Tissue Department of University Hospital Brno. The dataset consisted of data on blood demand from 2021 to 2024 and was extended by data regarding non-working days, national and school holidays, seasons, and influenza epidemics. The performance of various prediction models was measured using the normalized Mean Absolute Error (nMAE), which reflects the average prediction error relative to the average daily blood demand. When tested on data from 2023, the best performance was achieved by linear regression models, with a nMAE of 26% for A+ and 50% for O-, indicating lower predictability for blood types with smaller populations. Interestingly, models for different blood types use different features, as the demand for individual blood types depends on different factors. Despite relatively high nMAE values, the models still outperformed a”qualified guess” approach based only on historical averages.

Blood demand | computational modeling | feature selection | machine learning

Blood demand | computational modeling | feature selection | machine learning

DAŇKOVÁ, M.; KOSKOVÁ, S.; PLEŠINGER, F.

01.01.2025

Brno University of Technology

9788021463202

Proceedings II of the Conference Student Eeict

Proceedings II of the Conference STUDENT EEICT

Česká republika

72

75

4