Questions about comparing algorithms.

ReM

I'm writing my bachelor thesis on PM Algorithms. For that, I want to benchmark them on the same training/test set and compare their quality dimensions. Is there a way to get comparable values for each discovery method? E.g. is it possible to make quality assertions about fitness, when one algorithm outputs a Petri Net and another a fuzzy model? If so, any pointers how I would best go about it? Should I convert everything to a petri net, or can I use different fitness methods for each output model?

The methods I want to focus on are Fuzzy, Alpha, Heuristic, Multi-phase, Genetic, Region-based Miner. Are there any (new or old) important ones missing?

Lastly, is there a way to see who wrote which algorithm (or plugin) and when? I want to find the corresponding papers, but it has been very difficult so far. For instance, I couldn't find anything about Alpha-R, which I presume to be the latest Alpha Miner.

Thank you for your help.

sjjleemans

Fuzzy models do not have clear well-defined semantics, so they cannot be translated to anything else. The "easiest" is to translate all models to Petri nets. In http://www.sciencedirect.com/science/article/pii/S0306437912000464, many quality measures are discussed, which might give inspiration for your comparison.

A few other miners you could consider are Split Miner (A. Augusto), Fodina (S. vanden Broucke), Inductive Miner (S. Leemans), Evolutionary Tree Miner (J. Buijs), Integer Linear Programming Miner (J. van der Werf) etc. Via Google Scholar, you can easily find their defining papers.

Good luck with your thesis!

ReM

Thank you for your answer, I will look into those algorithms.

I hope ProM can translate into Petri Net for all necessary models.