Research On Process-aware Business Process Prediction Method Based On Deep Learning

Business process prediction is to analyze the event logs of recorded processes and predict the ongoing business processes in business process management.In recent years,deep learning has been increasingly used in business process prediction,and the mainstream approach is to treat processes as sequences in natural language processing,and use recurrent neural networks and LSTM to learn the semantics of processes like the semantics of word sequences.These methods are limited by the shortcomings of RNNs and may not to make good use of the behavioral relationships between activities in the process and the graph features of the process.so this paper proposes a process-aware business process prediction based on deep learning method,and the main research contents are as follows:(1)In order to address the shortcomings of business process prediction methods using RNNs and CNNs,and the failure to make good use of the behavioral relationships between activities in the process,this paper proposes a business process prediction method based on temporal convolutional networks(TCN),using TCN as the backbone network model for deep learning business process prediction and encode event sequences by behavioral vectors from the behavioral profile of business processes.And fusing the activity features of events with the attribute features of events in business processes for predicting the next event of business processes and generate the remaining processes by beam search based on behavioral profile of event logs.Experimental results on five real-life event log datasets show that TCN-based and behavioral vectors encoding business process prediction methods peformance improved on predicting next event accuracy and remaining process similarity compared to multiple methods.(2)Since the process naturally contains graph features,and graph neural networks(GNNs)is a deep learning technique proposed for the learning of graph-structured data.In this paper,we propose a method that represents business processes as graph-structured data,activities as nodes of the graph,and event attributes recorded in event logs as node features of the graph,and uses GAT to learn graph-structured representations of business processes to predict the duration of business processes.Experimental results on four real-life event log datasets show that the method using GNNs to predict business processes outperforms LSTM on process graphs with more nodes and is similar to LSTM on process graphs with fewer nodes,and GAT outperform GCN.Figure [13] Table [6] Reference [69]...