Continual learning, known also as lifelong learning, aims to design learning models that can continuously and autonomously adapt to varying data concepts without forgetting previously collected knowledge. Predictive business process monitoring, which predicts future process steps, is crucial in dynamic environments where tasks are not previously specified and processes frequently change or face unpredictability. However, many existing frameworks assume a static setting, ignoring the dynamic nature and concept drifts in processes, leading to catastrophic forgetting, where training over new data adversely affects the performance on previously learned tasks. This paper presents TFCLPM, a framework for online next activity prediction that operates without relying on predefined tasks and employs continual learning techniques to reduce catastrophic forgetting. The methodology combines a Single Dense Layer neural network with a continual learning algorithm to retain challenging historical samples and stabilize model parameters. Additionally, our approach introduces a diversity-aware memory updating strategy, ensuring that retained samples remain representative of the overall data distribution, leading to more stable adaptation and improved long-term retention. The proposed framework’s performance is compared against six existing online next activity prediction methodologies on synthetic and real-world event logs. Results show significant improvements in prediction accuracy, highlighting the framework’s robustness and efficiency, even on larger datasets with multiple classes.
Straten et al. (Fri,) studied this question.