TY - GEN
T1 - A Temporal Fusion Transformer for Long-Term Explainable Prediction of Emergency Department Overcrowding
AU - Caldas, Francisco M.
AU - Soares, Cláudia
N1 - Funding Information:
This work was partially supported by the strategic project NOVA LINCS (UIDB/04516/2020), the FCT project DSAIPA/AI/0087/2018 and the Carnegie Mellon University - Portugal FCT project CMU/TIC/0016/2021.
Publisher Copyright:
© 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2023
Y1 - 2023
N2 - Emergency Departments (EDs) are a fundamental element of the Portuguese National Health Service, serving as an entry point for users with diverse and very serious medical problems. Due to the inherent characteristics of the ED, forecasting the number of patients using the services is particularly challenging. And a mismatch between the affluence and the number of medical professionals can lead to a decrease in the quality of the services provided and create problems that have repercussions for the entire hospital, with the requisition of health care workers from other departments and the postponement of surgeries. ED overcrowding is driven, in part, by non-urgent patients, that resort to emergency services despite not having a medical emergency and which represent almost half of the total number of daily patients. This paper describes a novel deep learning architecture, the Temporal Fusion Transformer, that uses calendar and time-series covariates to forecast prediction intervals and point predictions for a 4 week period. We have concluded that patient volume can be forecasted with a Mean Absolute Percentage Error (MAPE) of 5.91% for Portugal’s Health Regional Areas (HRA) and a Root Mean Squared Error (RMSE) of 84.4102 people/day. The paper shows empirical evidence supporting the use of a multivariate approach with static and time-series covariates while surpassing other models commonly found in the literature.
AB - Emergency Departments (EDs) are a fundamental element of the Portuguese National Health Service, serving as an entry point for users with diverse and very serious medical problems. Due to the inherent characteristics of the ED, forecasting the number of patients using the services is particularly challenging. And a mismatch between the affluence and the number of medical professionals can lead to a decrease in the quality of the services provided and create problems that have repercussions for the entire hospital, with the requisition of health care workers from other departments and the postponement of surgeries. ED overcrowding is driven, in part, by non-urgent patients, that resort to emergency services despite not having a medical emergency and which represent almost half of the total number of daily patients. This paper describes a novel deep learning architecture, the Temporal Fusion Transformer, that uses calendar and time-series covariates to forecast prediction intervals and point predictions for a 4 week period. We have concluded that patient volume can be forecasted with a Mean Absolute Percentage Error (MAPE) of 5.91% for Portugal’s Health Regional Areas (HRA) and a Root Mean Squared Error (RMSE) of 84.4102 people/day. The paper shows empirical evidence supporting the use of a multivariate approach with static and time-series covariates while surpassing other models commonly found in the literature.
KW - Emergency department
KW - Explainable ML
KW - Forecasting
KW - Machine learning
KW - Manchester triage system
KW - National Health Service
KW - Neural network
KW - Temporal fusion transformer
KW - Time series
UR - http://www.scopus.com/inward/record.url?scp=85149839087&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-23618-1_5
DO - 10.1007/978-3-031-23618-1_5
M3 - Conference contribution
AN - SCOPUS:85149839087
SN - 978-3-031-23617-4
T3 - Communications in Computer and Information Science
SP - 71
EP - 88
BT - Machine Learning and Principles and Practice of Knowledge Discovery in Databases
A2 - Koprinska, Irena
A2 - Mignone, Paolo
A2 - Guidotti, Riccardo
A2 - Jaroszewicz, Szymon
A2 - Fröning, Holger
A2 - Gullo, Francesco
A2 - Ferreira, Pedro M.
A2 - Roqueiro, Damian
A2 - Ceddia, Gaia
A2 - Nowaczyk, Slawomir
A2 - Gama, João
A2 - Ribeiro, Rita
A2 - Gavaldà, Ricard
A2 - Masciari, Elio
A2 - Ras, Zbigniew
A2 - Ritacco, Ettore
A2 - Naretto, Francesca
A2 - Theissler, Andreas
A2 - Biecek, Przemyslaw
A2 - Verbeke, Wouter
A2 - Schiele, Gregor
A2 - Pernkopf, Franz
A2 - Blott, Michaela
A2 - Bordino, Ilaria
A2 - Danesi, Ivan Luciano
A2 - Ponti, Giovanni
A2 - Severini, Lorenzo
A2 - Appice, Annalisa
A2 - Andresini, Giuseppina
A2 - Medeiros, Ibéria
A2 - Graça, Guilherme
A2 - Cooper, Lee
A2 - Ghazaleh, Naghmeh
A2 - Richiardi, Jonas
A2 - Saldana, Diego
A2 - Sechidis, Konstantinos
A2 - Canakoglu, Arif
A2 - Pido, Sara
A2 - Pinoli, Pietro
A2 - Bifet, Albert
A2 - Pashami, Sepideh
PB - Springer
CY - Cham
T2 - Workshops on SoGood, NFMCP, XKDD, UMOD, ITEM, MIDAS, MLCS, MLBEM, PharML, DALS, IoT-PdM 2022, held in conjunction with the 21st Joint European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2022
Y2 - 19 September 2022 through 23 September 2022
ER -