Preweaning mortality remains a major constraint to productivity in hyperprolific sow herds. We investigated whether birth weight (BW), subcutaneous fat thickness, and management factors could improve early identification of high-risk piglets. A total of 5,967 live-born piglets from 473 Landrace × Yorkshire sows were followed from birth to weaning under commercial tropical production conditions. Daily mortality identified the first three days after birth as the critical period, accounting for approximately 45% of all preweaning deaths. Decision tree analysis identified biologically meaningful thresholds for BW (0.73 kg) and belly fat thickness (BFT; 0.50 mm), defining three risk groups: normal birth weight (NBW), low birth weight with high BFT (LBW–HBT), and low birth weight with low BFT (LBW–LBT). Compared with NBW piglets, mixed-effects Cox proportional hazards models showed markedly greater mortality hazards during the critical period for LBW–HBT (hazard ratio [HR] = 10.42; P < 0.001) and LBW–LBT piglets (HR = 23.26; P < 0.001). Among piglets surviving beyond this period, mortality remained approximately 3.5-fold higher in both low-birth-weight groups. Cross-fostering reduced early mortality, whereas non-cross-fostered piglets had a 38% greater hazard of death (P = 0.007). Among seven machine learning algorithms, Random Forest achieved the highest predictive performance (F1-score = 86.7%; area under the receiver operating characteristic curve = 0.777), with BW emerging as the most influential predictor. Integrating BW and BFT enables robust risk stratification at birth and provides a practical framework for precision livestock management through targeted interventions during the narrow window when most preweaning deaths occur.