Hurricane preparedness saves lives and reduces economic impacts, especially in United States Gulf Coast states, where climate change is intensifying disaster risk. Although risk is a key correlate of preparedness, most research has focused on subjective measures and rarely incorporates objective risk indicators, limiting the effectiveness of targeted preparedness efforts. To address this gap, survey data from a probability-based sample from the United States Gulf Coast and adjacent regions (N=2,135, response rate=78.4%, data collection dates=4/30/25-5/19/25) was merged with FEMA data on annualized frequency of hurricanes. K-means clustering classified respondents based on preparedness and the annualized frequency of hurricanes at their location. Three groups were identified: Low-Risk/Low-Preparedness (n=335, 16%), High-Risk/High-Preparedness (n=796, 38%), and High-Risk/Low-Preparedness (n=982, 46%). Using demographic and climate-exposure metrics, four machine learning algorithms (random forest, support vector machines, elastic-net logistic regression, and gradient boosting) predicted preparedness-risk cluster memberships with the highest accuracy of 66%. Consistent with machine learning findings, multinominal logistic regression found older age, identifying as a Black person, greater risk perception, higher number of disasters encountered, and higher hurricane exposure were associated with greater odds of belonging to the High-Risk/High-Preparedness group compared to the High-Risk/Low-Preparedness group. However, modest predictive accuracy suggests demographics and disaster- and climate-related factors do not fully explain hurricane preparedness in high-risk areas. We highlight prior methodological limitations, including the neglect of objective risk and overreliance on hypothesis significance testing, and discuss the need for more predictive approaches to advance the understanding of hurricane preparedness.