To address the limitations of early acute kidney injury (AKI) prediction, researchers have increasingly turned to machine learning methods. However, the success of these models hinges on the selection of relevant features. To this end, diverse feature selection techniques are employed to improve model generalization, stability, and interpretability.

Artificial intelligence (AI) has demonstrated promise for time-sensitive applications in AKI. These applications encompass early identification, warning, and the provision of AKI treatment recommendations. Machine learning-based models can detect AKI at an early stage, providing clinicians with a chance to intervene earlier and potentially improve patient outcomes.

In this study, we introduce a machine learning model that employing a two-stage feature selection process to predict in-hospital mortality risk among ICU patients with AKI. Our aim is to identify crucial features for mortality prediction and, as a result, reduce feature dimensionality to enhance model interpretability without sacrificing accuracy.

Read More