Predicting Unmet Healthcare Needs in Post-Disaster: A Machine Learning Approach.

Abstract

Unmet healthcare needs in the aftermath of disasters can significantly impede recovery efforts and exacerbate health disparities among the affected communities. This study aims to assess and predict such needs, develop an accurate predictive model, and identify the key influencing factors. Data from the 2017 Long-term Survey on the Change of Life of Disaster Victims in South Korea were analyzed using machine learning techniques, including logistic regression, C5.0 tree-based model, and random forest. The features were selected based on Andersen's health behavior model and disaster-related factors. Among 1659 participants, 31.5% experienced unmet healthcare needs after a disaster. The random forest algorithm exhibited the best performance in terms of precision, accuracy, Under the Receiver Operating Characteristic (AUC-ROC), and F-1 scores. Subjective health status, disaster-related diseases or injuries, and residential area have emerged as crucial factors predicting unmet healthcare needs. These findings emphasize the vulnerability of disaster-affected populations and highlight the value of machine learning in post-disaster management policies for decision-making.