Probabilistic forecasting improves timing predictions in business processes
Probabilistic Forecasting of Business Process Executions with Neural Temporal Point Processes
Machine Learning
Summary
Predicting what happens next in business workflows helps companies plan better, but knowing how reliable these predictions are is crucial. The authors developed a way to make forecasts that not only guess the next action but also provide a sense of uncertainty about when it will happen. Their method uses a special type of model that treats actions as events happening over time and can handle situations where multiple actions occur simultaneously. Tested on real data, their approach matched or outperformed existing methods in accuracy and gave better estimates of timing uncertainty, all while being computationally efficient.
What this means in practice
- •For business process analysts: Use probabilistic forecasts to assess the reliability of predicted next steps and timings in ongoing business workflows.
- •For service operations teams: Improve resource allocation by accessing reliable distributions of expected upcoming business activities and their timings.
Authors
Jiaxin Yuan, Daniela Grigori, Han van der Aa
Abstract
Operators of service-based systems act on forecasts of how a running execution will continue, and such a forecast is actionable only if its reliability is known. Mainstream deep-learning models for this task are discriminative and deterministic: they emit a single next activity and a single remaining-time estimate, without a distribution to reason over. We instead cast the problem as generative sequence modelling with marked temporal point processes, which define a joint density over the next mark and its inter-event time and therefore deliver predictive distributions by construction. Real event logs violate the simple-point-process assumption these models rest on, since consecutive events frequently carry identical timestamps; we handle such ties explicitly and combine a transformer encoder with a mixture decoder over inter-event times, trained by exact log-likelihood. On ten public logs, the resulting model matches discriminative baselines on point accuracy, dominates them on the calibration and sharpness of remaining-time distributions, and is the cheapest at inference, since a full predictive distribution is obtained in a single forward pass without sampling.