Analyzing and Predicting Verification of Data-Aware Process Models–A Case Study With Spectrum Auctions

Abstract

Verification techniques play an essential role in detecting undesirable behaviors in many applications like spectrum auctions. By verifying an auction design, one can detect the least favorable outcomes, e.g., the lowest revenue of an auctioneer. However, verification may be infeasible in practice, given the vast size of the state space on the one hand and the large number of properties to be verified on the other hand. To overcome this challenge, we leverage machine-learning techniques. In particular, we create a dataset by verifying properties of a spectrum auction first. Second, we use this dataset to analyze and predict outcomes of the auction and characteristics of the verification procedure. To evaluate the usefulness of machine learning in the given scenario, we consider prediction quality and feature importance. In our experiments, we observe that prediction models can capture relationships in our dataset well, though one needs to be careful to obtain a representative and sufficiently large training dataset. While the focus of this article is on a specific verification scenario, our analysis approach is general and can be adapted to other domains.