Machine learning-based FEMA Transitional Shelter Assistance (TSA) eligibility prediction models

Around 90 percent of the natural disasters in the United States involve floods. As a result of these floods, a massive number of houses become uninhabitable for their residents, making them in immediate need of lodging and shelters. The Federal Emergency Management Agency (FEMA) lodges people in noncongregated shelters such as hotels/motels for a short period-up to 45 days-through the Transitional Shelter Assistance (TSA) program. Government Accountability Office estimated that between 600 million and 1.4 billion dollars had been improperly spent. However, currently, the process of how an applicant becomes eligible for the TSA lacks a robust model and framework. However, the mechanism of selecting the recipients of TSA is mainly based on expert opinion and tacit knowledge. The objectives of this paper are (1) investigating how classification techniques can be used to help FEMA decision-makers during the time of the disaster and (2) building supervised machine learning decision-making models based on logistic regression, decision tree, and K nearest neighbor classification techniques using Python. The 4.8 million registries of applications dataset used for this paper were extracted from the National Emergency Management Information System. This research will help FEMA decision-makers for predicting TSA eligibility.