Optimizing extubation success: a comparative analysis of time series algorithms and activation functions

Area of Science:

• Critical Care Medicine
• Biomedical Engineering
• Data Science in Healthcare

Background:

• Extubation success is vital for patients with acute respiratory failure, as failure can lead to complications and patient distrust.
• Predicting extubation outcomes aids clinicians in managing mechanical ventilation and preventing adverse events.

Purpose of the Study:

• To compare various time series algorithms and activation functions for predicting extubation success or failure.
• To develop a reliable model for clinical decision support in the extubation process.

Main Methods:

• Comparison of different time series algorithms.
• Evaluation of various activation functions within predictive models.
• Utilizing Holdout cross-validation for model performance assessment.

Main Results:

• The Gated Recurrent Unit (GRU) model combined with the Tanh activation function demonstrated superior predictive performance.
• An accuracy of 94.44% was achieved for predicting extubation success or failure using the GRU-Tanh model with Holdout cross-validation.

Conclusions:

• The GRU model with Tanh activation offers a promising approach for predicting extubation outcomes.
• This prediction method can serve as valuable clinical decision support for healthcare professionals managing extubation.