Abstract Anemia is a prevalent public health problem associated with nutritional and socio-economic factors that contribute to iron deficiency. To understand the complex interplay of risk factors, we investigated a prospective population sample from the Jiangsu province in China. At baseline, three-day food intake was measured for 2849 individuals (20 to 87 years of age, mean age 47 ± 14, range 20–87 years, 64% women). At a five-year follow-up, anemia status was re-assessed for 1262 individuals. The dataset was split and age-matched to accommodate cross-sectional (n = 2526), prospective (n = 837), and subgroup designs (n = 1844). We applied a machine learning framework (self-organizing map) to define four subgroups. The first two subgroups were primarily from the less affluent North: the High Fibre subgroup had a higher iron intake (35 vs. 21 mg/day) and lower anemia incidence (10% vs. 25%) compared to the Low Vegetable subgroup. However, the predominantly Southern subgroups were surprising: the Low Fibre subgroup showed a lower anemia incidence (10% vs. 27%), yet also a lower iron intake (20 vs. 28 mg/day) compared to the High Rice subgroup. These results suggest that interventions and iron intake guidelines should be tailored to regional, nutritional, and socio-economic subgroups.