Federated Machine Learning in US Disaster Recovery Cost Prediction
Open Access DepositedThis research demonstrates how federated learning can be applied to disaster Recovery cost prediction across three agencies: the Federal Emergency Management Agency (FEMA), Housing and Urban Development (HUD), and Small Business Administration (SBA). In the absence of a machine learning solution in disaster Recovery cost prediction, the model training process compared three methods of cost: data-specific cost prediction, central cost prediction, and federated cost prediction. Nine separate model types were applied to the data at the individual and central level. It was determined that Extreme Gradient Boosting (XGBoost) was the best fit for the data. In the federated model, XGBoost was applied to each of the two partitions in the client level and Federated Averaging (FedAvg) was applied at the server level. The results demonstrate the feasibility of federated learning as a tool in disaster Recovery cost prediction, while maintaining privacy preserving needs of major legislation required in federal data sharing. The results proposed a scalable model that can be used in time-sensitive disaster scenarios for data-driven decision making where a central model is not available due to administrative delay. This approach is novel in the idea that machine learning can be used for cost prediction and federated learning can alleviate impediments in data sharing in the federal government.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.