Background: Perinatal arterial ischemic stroke (PAIS) is a leading cause of hemiparetic cerebral palsy. Multiple risk factors are associated with PAIS but studies are limited by small sample sizes and complex interactions. Unbiased machine learning applied to larger datasets may enable the development of robust predictive models. We aimed to use machine learning to identify risk factors predictive of PAIS and compare these to the existing literature. Methods: Common data elements of maternal, delivery, and neonatal factors were collected from three perinatal stroke registries and one control sample over a 7-year period. Inclusion criteria were MRI-confirmed PAIS, term birth, and idiopathic etiology. Random forest machine learning in combination with feature selection was used to develop a predictive model of PAIS. Results: Total of 2571 neonates were included (527 cases, 2044 controls). Risk factors uniquely identified through machine learning were infertility, miscarriage, primigravida, and meconium. When compared, factors identified through both literature-based selection and machine learning included maternal age, fetal tobacco exposure, intrapartum fever, and low 5-minute APGAR. Conclusions: Machine learning offers a novel, less biased method to identify PAIS predictors and complex pathophysiology. Our findings support known associations with concepts of placental disease and difficult fetal transition and may support early screening for PAIS.