Empirical Insights into Context-Aware Process Predictions: Model Selection and Context Integration

The prediction of service process performance is fraught with challenges from the selection of suitable recurrent neural network architectures. In this study, we evaluate the effectiveness of various recurrent neural network (RNN) architectures, including Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks, for predictive process monitoring in service processes. We analyze the impact of incorporating contextual process information from event logs by benchmarking the model performance across multiple IT service management datasets, RNN architectures, and integration strategies. Our empirical results highlight that context-aware models improve prediction performance, but the extent varies with the dataset and the specific RNN architecture. Time-aware LSTMs demonstrated superior performance for remaining time and next timestamp predictions, while GRUs were particularly effective in predicting the next activity. This research underscores the need to carefully consider model architecture and context integration to enhance predictive process monitoring. It also provides insights that can guide the selection of models and integration techniques, leading to future research directions in refining these methods for service process applications.