The goal of the retrospective study was to use random forest (RF) models to identify risk factors for the dissemination of carbapenem-resistant Klebsiella pneumoniae (CRKP) in neonatal units with a focus on clonal spread, healthcare groups (HGs), and plasmid dynamics. The study analyzed 64 CRKP isolates from 58 neonates between 2013 to 2020 at Shanghai Children’s Medical center that were less than 28 days old, had positive K. pneumoniae cultures, and isolates resistant to at least one carbapenem. Two RF models were used in the study. The first model used bacterial genotypes, plasmids, healthcare groups, and wards as independent variables and the time between isolates as the dependent variable to ultimately identify factors contributing to the persistence of CRKP isolates over time. The second model predicted genotypes associated with outbreaks of CRKP based on the presence of accessory genes. The study was able to identify ST433 and ST14 as responsible for three CRKP clonal outbreaks involving five or more neonates. The model was able to predict outbreak associated genotypes with 100% accuracy, identifying 21 genes that corresponded to outbreak associated genotypes. HGs were found to be short term mediators of transmission while plasmids played a role in the persistence of specific CRKP isolates. Plasmids were not specific to particular STs, however pSCMC1 and pSCMC3 played a critical role in persistence with pSCMC1-1 associated with ST433 and ST14-C1 outbreaks, and pSCMC3 associated with an ST14-C2 outbreak. The model was able to explain 86% of the variance in isolation times. The use of the RF models in the study is promising but requires future multi-center testing with a greater number of isolates to provide greater insights into CRKP reservoirs and transmission pathways and increased generalizability. The study also focuses on neonatal units, making it hard to determine its generalizability to other locations of the hospital.