Objective: Preterm infants are vulnerable to neurodevelopmental impairments due to immature brain development. Early identification of developmental heterogeneity and key perinatal predictors is essential for timely intervention and improved longterm outcomes. Methods: Preterm infants born at Tongji Hospital from July 2022 to March 2025 were included. Clinical data were collected during hospitalization, and developmental follow-ups were conducted at 3, 6, 9, and 12 months corrected age. Group-Based Trajectory Modelling (GBTM) identified neurodevelopmental trajectories. Six machine-learning models were constructed to predict trajectory categories, with performance evaluated by ROC curves and SHAP interpretation. Results: A total of 509 infants who completed all follow-ups were analyzed. Two trajectories were identified: Catch-up (46.56%) and Stable (53.44%). The Catch-up group showed a marked increase in developmental quotient (DQ) from 87 .12 to 103.36, while the Stable group increased from 106.02 to 109.91 over 3–12 months. The Random Forest model performed best (training AUC 0.991; validation AUC 0.916) and was selected as the final model. SHAP revealed seven key predictors: gestational age, caregiver age, caregiver experience, intracranial haemorrhage, hyperbilirubinemia, birth length, and caregiver education. Higher gestational age, absence of complications, birth length 40–45 cm, and caregivers aged 25–35 years with prior experience and higher education were associated with the Stable trajectory . Conclusion: Preterm infants exhibit two neurodevelopmental patterns: Stable and Catch-up. Gestational age and caregiver characteristics strongly influence trajectories, emphasizing the importance of early risk stratification and family-centered developmental support.