From filler to pelleted seed germination: a data-driven approach using multiple linear regression models for seed optimisation

Seed pelleting improves the sowing precision of small, lightweight seeds; however, the formulation of optimal materials remains largely empirical. In this study, a data-driven framework was developed to quantify the relationships between filler properties, pelleted seed characteristics, and germination performance under varying humidity conditions. Fifteen fillers were evaluated using ten powder indicators, seven pellet traits and eight germination indices. Multiple linear regression (MLR) revealed that filler properties were stronger predictors of germination than pellet traits, with 23 of 24 filler-based models achieving R 2 > 0.6 compared to only 10 pellet-based models. Tap density and hygroscopicity emerged as dominant predictors across humidity regimes. Among the fillers, activated charcoal, zeolite and vermiculite performed best under low, medium and high humidity, respectively, whereas gypsum consistently underperformed. These results highlight the critical role of the filler responsiveness to environmental moisture. The MLR framework establishes a quantitative foundation for rational filler selection and provides a precise alternative to empirical approaches, paving the way for AI-assisted climate-adaptive seed pelleting strategies.