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

Area of Science:

• Disaster Management
• Applied Machine Learning
• Public Policy

Background:

• Floods cause widespread housing uninhabitability in the US, necessitating emergency shelter programs.
• The Federal Emergency Management Agency (FEMA) provides Transitional Shelter Assistance (TSA) for up to 45 days.
• Current TSA eligibility determination relies on expert opinion, lacking a robust, data-driven framework, leading to an estimated $600 million to $1.4 billion in improper spending.

Purpose of the Study:

• To investigate the application of classification techniques for FEMA disaster decision-making.
• To develop supervised machine learning models for predicting TSA eligibility.
• To provide FEMA with a data-driven tool for more accurate and efficient disaster relief allocation.

Main Methods:

• Utilized a 4.8 million record dataset from the National Emergency Management Information System.
• Implemented and compared logistic regression, decision tree, and K-nearest neighbor classification algorithms.
• Developed supervised machine learning models using Python for TSA eligibility prediction.

Main Results:

• Successfully built and evaluated machine learning models for predicting TSA eligibility.
• Demonstrated the potential of classification techniques to assist FEMA decision-makers.
• Identified specific algorithms (logistic regression, decision tree, K-nearest neighbor) as viable tools for this application.

Conclusions:

• Machine learning classification techniques offer a robust framework to improve FEMA's TSA eligibility determination process.
• The developed models can aid FEMA in making more informed decisions during disaster relief operations.
• Implementing these data-driven models can enhance the efficiency and accuracy of shelter assistance distribution, potentially reducing improper expenditures.