Data Science & Social Safety Net

This paper explores how data science can be used to strengthen the social safety net. As one such approach, the paper develops a model to predict the likelihood of hospital admission for Medicare and Medicaid insurers in the U.S. The analysis draws from the health insurance claims data of 45,000 patients with 939 features, spanning eight quarters from 2014 and 2015. Six models are adopted to predict patients' hospital admissions in 2016, based on their sociodemographic and health-related characteristics. The paper presents the rationale, processes, and results of analysis from logistic regression, Decision Tree, Random Forest, Support Vector Machine, feed-forward Multi-layer Perceptron (MLP), Convolution Neural Network, and Recurrent Neural Network. The best performing model, evaluated against recall and precision scores, is the MPL. This simple deep learning model was correct about 80% of the time for patient admissions to hospital. Additionally, tree-based algorithms provide important features related to hospital admission, such as medical risk scores. As a policy implication, the paper discusses predictive risk modeling to provide preventive care for at-risk populations. The paper concludes by suggesting strategies for using the data science approach in the allocation of social welfare programs and services.

​

..................................................................................................................................................................................