Time Matters: Time-Aware LSTMs for Predictive Business Process Monitoring

Abstract

Predictive business process monitoring (PBPM) aims to predict future process behavior during ongoing process executions based on event log data. Especially, techniques for the next activity and timestamp prediction can help to improve the performance of operational business processes. Recently, many PBPM solutions based on deep learning were proposed by researchers. Due to the sequential nature of event log data, a common choice is to apply recurrent neural networks with long short-term memory (LSTM) cells. We argue, that the elapsed time between events is informative. However, current PBPM techniques mainly use "vanilla" LSTM cells and hand-crafted time-related control flow features. To better model the time dependencies between events, we propose a new PBPM technique based on time-aware LSTM (T-LSTM) cells. T-LSTM cells incorporate the elapsed time between consecutive events inherently to adjust the cell memory. Furthermore, we introduce costsensitive learning to account for the common class imbalance in event logs. Our experiments on publicly available benchmark event logs indicate the effectiveness of the introduced techniques.

Modelling and Predictive Monitoring of Business Processes under Uncertainty with Reinforcement Learning

“…Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) have attracted increased research attention [ 20 , 21 , 22 , 23 , 24 ]. They have been used for predicting the next event of a running case and its timestamp [ 20 , 24 ]; predicting sequences of the next events and their associated resource pools [ 20 ]; predicting the remaining time [ 25 ]; and modelling the time dependencies between events [ 26 ]. Despite the promising results of deep learning methods in predictive business process monitoring, their explainability has arisen as a challenge [ 27 , 28 , 29 ], and they require vast amounts of data [ 30 ].…”

ProcessTransformer: Predictive Business Process Monitoring with Transformer Network

“…Likewise, the same process monitoring problems were addressed by Camargo et al [4] using the composition of LSTM to support both categorical and numeric features. Additional temporal features are proposed in [20] and [21] to improve the predictive capabilities of existing deep models. An extension of LSTM with attention mechanism is used in [38] for the process outcome prediction task.…”

“…Our approach has achieved the lowest MAE on average for all considered datasets compared to the previously proposed methods. Besides algorithmic specifications, our approach is different to [33,21,15,3] in following aspects. We deal with the event time prediction as an independent task as opposed to the common multi-task approach.…”

“…Evermann et al, [10] introduced recurrent neural networks (RNNs) for predicting the process behavior at run-time. Following this, several prediction models based on RNNs and related variants, such as long-short term memory (LSTM) networks have been proposed for the tasks of next activity [33,21,35], suffix generation [18,4], outcome prediction [36], and process's remaining time prediction [27,20]. Even though deep RNNs based PBPM methods have been firmly established for event sequence modeling, they suffer from considerable shortcomings, specifically for the next activity prediction task.…”

“…Firstly, several proposed methods [33,4,20,21,35] employ one-hot encoding to obtain the numeric representations of categorical events sequences. These integer representations disregard the intrinsic relationship among events and introduce unrealistic computational requirements due to an increase of data dimensionality [12].…”