The objective of this research project was to develop a predictive risk model for the release of hazardous material transported in unit trains using data science techniques for available rail accident and traffic data. This work builds upon previous research (Bing, et al., 2015) which examined the causal sequence of events which can lead to a rail accident and used historical accident record and rail traffic data to define conditional probabilities of occurrence and thereby predict the risk of a hazmat release. Three different Bayesian Networks (BNs) were implemented to study the causal relationships between weather, track, and train related risk factors and the primary causes leading to railroad accidents. The BN-based accident model has demonstrated potential to accurately predict the accident cause, given information about the train and track.