Time and Activity Sequence Prediction of Business Process Instances

Mirko Polato, Alessandro Sperduti, Andrea Burattin, Massimiliano de Leoni

Abstract

The ability to know in advance the trend of running process instances, with respect to different features, such as the expected completion time, would allow business managers to timely counteract to undesired situations, in order to prevent losses. Therefore, the ability to accurately predict future features of running business process instances would be a very helpful aid when managing processes, especially under service level agreement constraints. However, making such accurate forecasts is not easy: many factors may influence the predicted features. Many approaches have been proposed to cope with this problem but all of them assume that the underling process is stationary. However, in real cases this assumption is not always true. In this work we present new methods for predicting the remaining time of running cases. In particular we propose a method, assuming process stationarity, which outperforms the state-of-the-art and two other methods which are able to make predictions even with non-stationary processes. We also describe an approach able to predict the full sequence of activities that a running case is going to take. All these methods are extensively evaluated on two real case studies.

Introduction

In modern business process management, predicting the remaining time of a running process instance is critical for meeting Service Level Agreements (SLAs). Traditional process mining techniques often struggle with "non-stationary" processes—scenarios where the process logic evolves over time due to seasonal shifts, regulatory changes, or resource fluctuations. This paper addresses the challenge of providing accurate time and sequence forecasts even when a process instance follows a path not previously observed in historical logs.

Methodology

The core of the proposed approach lies in combining structural process models with machine learning. The input for these models is an event log, which records the execution of activities within various process instances.

Example of an event log used as input for process prediction models.

As shown in Figure 1, each event is associated with a Case ID, a timestamp, an activity, and various attributes like "Resource" or "Amount." The authors propose three primary methods:

  1. Simple Regression: Utilizes $\epsilon$-Support Vector Regression (SVR) to predict remaining time based on process attributes as input features.
  2. SVR with Contextual Information (SVR+TS): This method enhances SVR by incorporating structural information from a Transition System (TS). It introduces a similarity mechanism (e.g., Jaccard similarity or Damerau-Levenshtein distance) to map "non-fitting" traces to the most similar known states, ensuring robust predictions for unseen process variants.
  3. Data-aware Transition System (DATS): Each state in the TS is augmented with a Naive Bayes classifier for next-activity prediction and an SVR for transition-time prediction.

Furthermore, the DATS model is extended for Future Path Prediction (FPP). By transforming transition probabilities into additive "costs" via a negative log transformation, the problem of finding the most likely future activity sequence is solved using shortest-path graph algorithms like Best-First Search.

Experimental Evaluation

The methods were evaluated using two real-world datasets: a Road Fines Log (stable, linear) and a Help Desk Log (complex, non-linear).

Performance was measured using Mean Absolute Percentage Error (MAPE). On the Road Fines log, the DATS and SVR+TS methods reduced the error rate to approximately 2.8%, compared to a 5.89% baseline. In the more challenging Help Desk study, the SVR+TS method demonstrated superior robustness to non-stationarity; when process variants were intentionally withheld from training, SVR+TS achieved a MAPE of ~33.7%, significantly outperforming the baseline and DATS (~41%). Additionally, the FPP approach correctly identified the next activity with 94% accuracy.

Conclusion

By integrating structural transition systems with data-driven regression models and similarity-based mapping, this research provides a robust framework for business process prediction. The ability to handle non-stationary processes and provide accurate future path sequences offers a significant advantage for proactive process management. Future work will focus on refining hyperparameter tuning and incorporating human-centric factors like business hours and holidays into the temporal predictions.

Summary updated

Abstract

The ability to know in advance the trend of running process instances, with respect to different features, such as the expected completion time, would allow business managers to timely counteract to undesired situations, in order to prevent losses. Therefore, the ability to accurately predict future features of running business process instances would be a very helpful aid when managing processes, especially under service level agreement constraints. However, making such accurate forecasts is not easy: many factors may influence the predicted features. Many approaches have been proposed to cope with this problem but all of them assume that the underling process is stationary. However, in real cases this assumption is not always true. In this work we present new methods for predicting the remaining time of running cases. In particular we propose a method, assuming process stationarity, which outperforms the state-of-the-art and two other methods which are able to make predictions even with non-stationary processes. We also describe an approach able to predict the full sequence of activities that a running case is going to take. All these methods are extensively evaluated on two real case studies.
BibTeX
@article{polato2018time,
  author = {Mirko Polato and Alessandro Sperduti and Andrea Burattin and Massimiliano de Leoni},
  title = {{Time and Activity Sequence Prediction of Business Process Instances}},
  journal = {Computing},
  year = {2018},
  doi = {10.1007/s00607-018-0593-x},
}
Figure