Fig. 6From: Interpretable recurrent neural network models for dynamic prediction of the extubation failure risk in patients with invasive mechanical ventilation in the intensive care unitThe impacts of the top 20 features on predictions of LSTM_1. Each colored dot of the plot demonstrated the impact of a feature on the prediction for a 4-h window in the time sequence of a patient. The color bar represented the feature value, from low (blue) to high (red). The x-axis indicated the impacts on the model output, driving the prediction towards extubation failure (positive SHAP value) or towards extubation success (negative SHAP value). Abbreviations: GLU glucose, PEEP positive end expiratory pressure, Pmean mean airway pressure, Ppeak peak inspiratory pressure, P/F PaO2/FiO2Back to article page