This project provides estimates of the projected annual number of housing units for each of 1,000 counties in coastal states from Delaware to Texas over a 10-, 20-, and 30-year time horizon (2020-2029, 2020-2039, 2020-2049, respectively), using a long short-term memory (LSTM) neural network model. The absolute relative testing errors for the 10-, 20-, and 30-year projection models are 0.29%, 0.20%, and 0.25%, respectively. These housing inventory projections can be integrated with existing natural hazard risk models to evaluate the effects of a changing building inventory on regional risk. This project supports the publication by Williams et al. (2022) and contains the data and scripts used to evaluate three candidate modeling methods (linear trend models, autoregressive integrated moving average (ARIMA) models, and LSTM models) that predict the annual number of housing units per county. Python and R scripts including data processing steps, forecast modeling methods, results processing, and analysis figures are provided. Please see the “Williams et al. 2022_Supplemental Section.pdf” document for background information supporting the files contained in this project. Supporting Publication: Williams, C. J., R. A. Davidson, L. K. Nozick, J. E. Trainor, M. Millea, and J. L. Kruse.. (2022). “Regional county-level housing inventory predictions and the effects on hurricane risk.” Natural Hazards and Earth System Sciences.