Answers are grounded in this paper and cite their sources.
Machine learning models for predicting alcohol use relapse lack adequate explanations. This study aims to utilize visualization techniques to (1) highlight a machine learning model's predictive reasoning and biases; and (2) to identify the risk factors associated with readmission following inpatient treatment for alcohol use disorder. Our visualizations show that first, the model was reliable in its negative predictions but had difficulty in correctly identifying positive instances with missing variable values; second, a history of multiple treatments may distinguish otherwise similar instances; and third, highly impactful risk factors for 30-day readmission included a history of multiple treatments and lower participation in treatment. The visualizations when considered together could help clinical users understand model biases.
No takes yet. Share an insight, caveat, or question.
Liang et al. (2024) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: