Hi. And welcome to this lecture, conformance checking in ProM. So within the whole process mining spectrum, in this week we focus on conformance checking and extension. And in the last lecture, we discussed conformance checking using alignments, where we take the event log data and align it with the process model. And this process model can be discovered but you can also create it manually and put it in and compare it with the event log data. So using these alignments and this conformance checking, you can see where the data deviates from the model. So let’s go to ProM and show you how you can do this using real data and real models.
So with ProM open, let’s load the loan example process event log. And let’s discover a process model using the inductive miner to mine the Petri net with the inductive miner. And you get this process model, which looks like the one we’ve shown in the lecture. However in this data, there are some deviations. So let’s try and detect them by using this process model with the event log. And what we want to do is replay the log on a Petri net for conformance. So select this plug-in for now. And press Start. So in this wizard, you have to select how to map the activities in the process model with the events observed in the event log.
And in this case, we have a perfect match. So everything is OK. Now you can do some further settings. But in general, it’s OK to just press Next. And here you can even specify how much it would cost to do a move on model or a move on log on certain activities. So you can indicate preference for some deviations over others. Well, let’s click Finish. And now we see the same process model again but with colors. So let’s zoom in a little bit by scrolling the mouse wheel. And what we can see in the legend is that higher frequency activities get a darker blue color. So these activities have been executed more often.
And for instance, acceptance has only been executed 20 times. What you also see is that this activity has a red border and a green and purple bar on the bottom. And as is indicated by the legend, this indicates that 90 times the activity was executed synchronously or correctly and 10 times there was a model move. So in 10 out of 100 cases, this activity was not observed in the traces. So no check in the system was made. You can also see this by opening the Element statistics. And then you see indeed 90 times model move or synchronous move and 10 times a model move in 10 traces.
Usually you want to get a global number indicating how well the data aligns with the process model and, therefore, you look at the trace fitness. And in this case, this data complies for 99% with this model. So let’s do the same thing but now on real data. So let’s import the BPI Challenge 2012, only considering the A and O activities. So when we do this, we can again use the inductive miner to discover a Petri net.
And this is the Petri net we get. And now we can press the Play button, select the BPI Challenge Event Log. And again, do Replay of the event log on the Petri net for conformance analysis. Again, we have to indicate that these activities in the model relate to these activities in the event log. But we get a perfect mapping. And we keep all the settings as is. And now it is aligning all the traces and all the events on this process model. And now you, again, see the process model with the transitions colored in. They still mean the same thing.
And you actually see that only this one, this transition has deviations, but now you see also that certain places have different colors, they are actually yellow. That means that in this place or in actually in this marking so that there was a token in both places. In this situation in the process, we have observed certain move on logs. So for instance, 11 time when there was a token in these two places, 11 times we see, O CANCELED. And 35 times we’ve seen, O CREATED. Well, at this point in time, this was not as expected. And so you can click all the places and see what combined markings you have observed, what activities that you did not expect.
And if we look at the global statistics of this event log, we see that it has a trace fitness of almost 93%. Not that bad but still there are deviations that you might want to inspect. Another way to show deviations is by using the inductive visual miner. So let’s go back to the Object View, the BPI Challenge Event Log, and then run the Inductive Visual Miner. So let’s wait until it finds a model.
Put it on the concept name classifier.
And this is the process model that we get. And it automatically already starts aligning, as you can see in the bottom, and now it’s actually already creating the animation. So if we zoom in, we can see how the process is executed. And if we say show paths and deviations, you actually see some red arcs. And this indicates that this part has been skipped three times. And here for instance at this point in time, 200 times an activity that was not expected was observed. And this also helps you indicate where bottlenecks are, although it shows a bit less detail than the alignments that we’ve shown before.
So using ProM, we can show in several ways how the data aligns or misaligns with the process model.
So now that you’ve seen in ProM how you can do this, you know how you can validate where the data deviates from your process model. And now also that you can relate the data to the process model, we can also go towards extension of the process model. And that’s something we will discuss in the next lectures. Hope to see you again soon.