Predictive business process monitoring (PBPM) aims at predicting the future of running units in a process (prefixes), be it for the next event or the remaining sequence of events (suffix), an event being characterized at least by a triplet made of a unit identifier, an activity the unit can go through, and the time of activity execution. For suffix predictions, encoder-decoder generative adversarial networks (GAN) have proven to be most efficient. We improved the model further by using the Wasserstein loss (WGAN) to stabilize and ease its training phase. Our main contribution, however, comes with feature engineering: external, unobserved elements often have an impact on the progress of a process instance, and covariables are hardly ever taken into account in current models, or outright unavailable. We therefore propose a feature engineering method to extract what we call process crowding data, and predict suffixes conditionally to such variables by turning the model into a conditional WGAN. We will then show its efficiency at proposing noticeably improved predictions.
