In an example embodiment, a method for processing payments made via an electronic payment processing system is provided. An example method includes obtaining training data from a data source. The training data relates to prior purchases made via the electronic payment processing system, wherein the data source includes, in some examples, only a checkout page in a purchase transaction funnel. Features associated with a negative user action in relation to prior purchases are identified. A machine learning algorithm produces a dynamic transactional behavior score indicative of a probability that a purchase will invoke a negative user action.