Hi and welcome back. In this lecture, I will show you how you can use the fuzzy miner in ProM. For the fuzzy miner, it is the last algorithm we will discuss that discovers a process model from event data. The fuzzy miner, however, is a slightly different process discovery algorithm, since it does not discover Petri nets but process graphs. The problem with process graphs is that you cannot distinguish between choice and parallelism. So whether you do two activities in any order or one of the two. However, the fuzzy minor is really handy to quickly explore and get initial insights in your data. However, you can’t really get hard insights because you cannot decide whether something is a choice or parallel.
The fuzzy miner or derivatives thereof are used in commercial tools like Disco and Celonis, because they have a high practical value. And the implementation can be done quite efficiently. So let’s switch to ProM and show you how you can discover a fuzzy graph and what you can do with the fuzzy miner. So let’s import the loan example. Again, the partial event log. And let’s search for the fuzzy miner.
Mine for a Fuzzy model, and as you see, it results in a fuzzy model. And when you start this, the first question you get is which metrics you want to include. Automatically, it selects all, and usually, this is the right choice, but you can inspect some metrics further if you want to get some slightly different results. But usually, this is fine. Also, these settings are fine. But again, play around a little bit to see what the effects are if you want. And on this data, this is actually the process model that is discovered. But what we can do is reduce the node cutoff and you get all the possible combinations between directly follows of particular activities.
And I hope you can see that although the original data contains parallelism and choice, you cannot really distinguish that from this graph. And therefore, although it gives interesting insights on which particular orders activities can be executed, sometimes, you need a bit more clear process models. However, the fuzzy miner is able to handle large event logs as I will show you now. So let’s import the BPI 2012 challenge event log. And let’s again filter it to only contain the A and O activities.
Let’s keep all end activities. And select the A and the O. And let’s rename the event log.
So now, we have only the A and O activities. And again, let’s search for the Fuzzy miner. And again, the initial settings and the wizard screens are OK. And rather quickly, it comes up with a model. However, for instance, what you see is that some activities are disconnected. So there is no clear relationship on where they fit in the main process. But however, you can play with the nodes. So for instance, if you move the node filter up, more and more activities are combined into a cluster. In the extreme case, you have this. But when you move to the bottom, all activities remain.
And similarly, as I’ve shown you before, you can remove many edges or you can include very many edges. And the thickness and the darkness of the edge indicate how frequent it is observed. What you can also do is you can say that this is the model you want, we can right click and export the Fuzzy model. As you go back to the object view, you have Fuzzy model, and this one, together the original event log– so now I’ve selected the event log and the Fuzzy graph that I exported. Now we can animate the event log on the Fuzzy instance.
So if you press Finish, it takes a while to animate and prepare everything, but now we see the same process of the same Fuzzy graph. And when we press play, we see how the cases, so a token is a particular case, how they float through the process model. And on the left hand side, you see how quickly they follow particular phases of the process. And you can see that currently we have completed 8% or 9% of all the cases in the event log. At any point in time, you can pause it and you see the actual date and time in reality that you’re currently looking at.
So similar to the inductive miner, you also have visualization, and you get some details, but the graph structure is sometimes hard to interpret.
So now you know how you can use the Fuzzy miner in ProM. And as I’ve shown in the screencast, this is the result of the Fuzzy miner on the example data that we used throughout this course. And when we look at the check list, there’s not much you can say about soundness and replay fitness. This is not really a Petri net, but a process graph. So therefore, it’s very handy to get quick insights and to see what the main structure of a process is, but not so much for really validating rules and conformance checking. So this concludes the process discovery part of this course. And now we know how we can discover a process model.
We can do performance checking and enhancement. I hope to see you again in the next lectures.