The global incidence of infant food allergy is rising, yet effective primary prevention strategies remain elusive. Identifying key early-life risk factors is critical for prevention, but the highdimensional and complex nature of these exposures poses a significant challenge for onventional statistical methods. To address this gap, this study proposes a novel feature selection method based on a Grey Wolf Optimizer (GWO). By incorporating a non-linear control parameter, an adaptive diversity-driven mechanism, and a cryptobiosis-inspired strategy , the algorithm effectively overcomes the limitations of the standard GWO. The superiority of EGWO was demonstrated against other state-of-the-art algorithms on the CEC 2017 benchmark functions. We applied the binary version of this method to a prospective birth cohort of 484 mother-infant pairs. The analysis successfully identified a minimal yet highly predictive set of five key risk factors: gestational diabetes, frequent maternal alcohol consumption during pregnancy , infant azithromycin use within six months, and the pre-pregnancy BMI of both parents. These findings provide core evidence for the etiology of food allergy , pointing to key regulatory pathways in early-life immune development. Our work highlights the utility of advanced computational methods for pinpointing crucial biomarkers and provides a datadriven foundation for precise, preconception-initiated prevention strategies.