Synthetic Training-Data Generation for ML-based Process Mining Tools

This work addresses the challenge of data scarcity in process mining by proposing the creation of synthetic training data using generative models. A comparative analysis is conducted between a Long Short-Term Memory (LSTM) model and the Generative Adversarial Network (GAN) model, using two distinct datasets. Multiple evaluation methods are employed to compare the results from the two models based on: precision, fidelity, diversity, and novelty. Results indicate that while LSTM accurately reproduces the initial data structure, GAN introduces more variability, offering a wider range of training scenarios. This highlights the potential of GAN-generated data to enhance the effectiveness and reliability of machine learning-based process mining tools