International Conference on Computer and Knowledge Engineering

Home / 14th International Conference on Computer and Knowledge Engineering

An Attention-Based Model for Clinical Time Series Prediction: Enhancing ICU Readmission Prediction

Authors :

Hananeh Sadat Madinei¹ Mohammad Reza Keyvanpour² Seyed Vahab Shojaedini³

1- Department of Computer Engineering, Alzahra University 2- Department of Computer Engineering, Alzahra University 3- Department of Electrical Engineering, Iranian Research Organization for Science and Technology

Keywords :

clinical time series،electronic health records،readmission prediction،deep learning،attention mechanism،positional encoding

Abstract :

In recent years, time series prediction has become a highly interesting topic in various applied areas, including clinical time series analysis. Hospitals and other clinical healthcare systems collect Electronic Health Records. These records contain important information that needs to be mined. This information provides a solid basis for predicting hospital readmission, helping to improve healthcare, enhance patient outcomes, and optimize hospital resources. This study introduces a novel deep learning attention-based approach for predicting 30-day ICU readmissions using clinical time series information extracted from the MIMIC III dataset. This approach incorporates time-aware positional encoding into the attention mechanism to enhance predictive performance by capturing the temporal representation of data and considering the temporal order of clinical time series data. This model is compared with cutting-edge models. The significant enhancements in the model performance achieved by the proposed method highlight the effectiveness of incorporating time-aware positional encoding in capturing temporal dependencies and improving prediction outcomes. These results underscore the potential of advanced attention mechanisms in clinical time series analysis.