Abstract Next activity prediction of business processes (BPs) provides valid execution information of ongoing (i.e., unfinished) process instances, which enables process executors to rationally allocate resources and detect process deviations in advance. Current researches on next activity prediction, however, concentrate mostly on model construction without in‐depth analysis of historical event logs. In this article, we are dedicated to proposing an approach to forecast the next activity effectively in BPs. After in‐depth analysis of historical event logs, three types of candidate activity attributes are defined and calculated as additional input for the prediction based on three essential elements, that is, frequent activity patterns, trace similarity and position information. Furthermore, we construct an effective hybrid prediction model combining the popular convolutional neural network (CNN) and bidirectional long short‐term memory (Bi‐LSTM) with self‐attention mechanism. Specifically, CNN is used to extract the temporal features before importing into Bi‐LSTM for accurate prediction, and self‐attention mechanism is applied to strengthen features that have decisive effects on the prediction results. Comparison experiments on four real‐life datasets demonstrate that our hybrid model with selected attributes achieves better performance on next activity prediction than single models, and improves the prediction accuracy by 2.98%, 6.05%, 2.70% and 5.26% on Helpdesk, Sepsis, BPIC2013 Incidents and BPIC2012O datasets than the state‐of‐the‐art methods, respectively.

Process discovery techniques analyze process logs to extract models that characterize the behavior of business processes. In real-life logs, however, noises exist and adversely affect the extraction and thus decrease the understandability of discovered models. In this paper, we propose a novel double granularity filtering method, executed on both the event and trace levels, to detect noises by analyzing the directly-following and parallel relations between events. Based on the probability of an event occurring in a sequence, the infrequent behaviors and redundant events in the logs can be filtered out. In addition, the missing events in parallel blocks are detected to further improve the performance of filtering. Experiments on synthetic logs and five real-life datasets demonstrate that our method significantly outperforms other state-of-the-art methods.

Remaining activity sequence prediction (i.e., Activity suffix prediction) aims at recommending the most likely future behaviors for ongoing process instances (traces), which enables process managers to rationally allocate resources and detect process deviations in advance.Recently, techniques of neural networks have found promising applications in activity suffix prediction by training a prediction model for next activity and iteratively performing the model to achieve the whole sequence prediction.However, the iterative prediction accumulates the deviations of each iteration and the result also lacks interpretability.In this paper, we propose a novel method to predict activity suffixes from the perspective of control flow and data flow for ongoing traces, where process discovery and trace replay techniques are employed to simulate executions of traces under real conditions and Long Short-Term Memory (LSTM) is applied to characterize the correlation between executed information and future execution.Sequence matching between historical prefix traces and ongoing traces are performed based on the above information to select the optimal-matched (i.e., most similar) activity suffix for ongoing process instances.Experiments on real-life datasets demonstrates that our proposed method outperforms other methods.

Remaining activity sequence prediction (i.e. activity suffix prediction) aims at recommending the most likely future behaviors for ongoing process instances (i.e. traces), which enables process managers to rationally allocate resources and detect process deviations in advance. Recently, techniques of neural networks have found promising applications in activity suffix prediction by training a prediction model for next activity and iteratively performing the model to achieve the whole sequence prediction. However, the iterative prediction accumulates the deviations of each iteration and the result also lacks interpretability. In this paper, we propose a novel method to predict activity suffixes from the perspective of control flow and data flow for ongoing traces, where process discovery and trace replay techniques are employed to simulate executions of traces under real conditions and Long Short-Term Memory (LSTM) is applied to characterize the correlation between executed information and future execution. Sequence matching between historical prefix traces and ongoing traces is performed based on the above information to select the optimal-matched (i.e. most similar) activity suffix for ongoing process instances. Experiments on real-life datasets demonstrate that the proposed method outperforms other methods.

Remaining time predictive monitoring of business processes (BPs) is a key research issue in business process mining, which provides timely predictive information for stakeholders to take proactive corrective actions to reduce process execution risk such as exceeding time limit or to adjust the priority of activities. However, current researches on remaining time prediction only consider the impact of internal attributes of single process instance, but ignore the resource competition among multiple instances executed together. Therefore, this paper takes resource competition into consideration and characterizes several inter-instance attributes as the input of prediction. We also prioritize and select some key activities that strongly impact the execution time of BPs according to historical event logs and include them as input of the prediction. Meanwhile, in order to solve the instability of one single prediction model in complex scenarios, a multilayer hybrid model constructed from XGBoost and LightGBM models using stacking technique is proposed. Experiments on four real-life datasets show that our approach of considering attributes among instances and including key activities into a hybrid model outperforms other prediction methods.