Drought stress is a major constraint on crop development, potentially causing huge yield losses and threatening global food security. Improving Crop's stress tolerance is usually associated with a yield penalty. One way to balance yield and stress tolerance is modification specific gene by emerging precision genome editing technology. However, our knowledge of yield-related drought-tolerant genes is still limited. Foxtail millet (Setaria italica) has a remarkable tolerance to drought and is considered to be a model C4 crop that is easy to engineer. Here, we have identified 46 drought-responsive candidate genes by performing a machine learning-based transcriptome study on two drought-tolerant and two drought-sensitive foxtail millet cultivars. A total of 12 important drought-responsive genes were screened out by principal component analysis and confirmed experimentally by qPCR. Significantly, by investigating the haplotype of these genes based on 1844 germplasm resources, we found two genes (Seita.5G251300 and Seita.8G036300) exhibiting drought-tolerant haplotypes that possess an apparent advantage in 1000 grain weight and main panicle grain weight without penalty in grain weight per plant. These results demonstrate the potential of Seita.5G251300 and Seita.8G036300 for breeding drought-tolerant high-yielding foxtail millet. It provides important insights for the breeding of drought-tolerant high-yielding crop cultivars through genetic manipulation technology.