Nutritionally vulnerable populations struggle to maintain basic health and quality of life because they have difficulty purchasing and consuming enough of a variety of foods. Efficient targeting methods are needed to support the undernourished, but research is lacking. The purpose of this study is to analyze the impact of information diversity on the performance of prediction models for the undernourished and suggest ways to improve the efficiency of targeting. Using the 2021 National Health and Nutrition Examination Survey and machine learning methodology, we proposed an optimal model for identifying nutritionally vulnerable people and suggested important variables that contribute to improving the prediction ability. The results of the study showed that adding information such as health status, health behavior, and food intake features to individual characteristics improves the predictive ability of the machine model. A strategy to collect relevant information and use it to identify nutritionally vulnerable people is necessary. In addition, identifying and utilizing information that contribute to the predictive power of food security by sample can help develop strategies to proactively identify and quickly support nutritionally vulnerable populations.