Hi, and welcome to the closing lecture of week 3. In this week, we discovered additional process mining activities and, of course, also, how you can evaluate them and summarize their results. And in this week, we did not focus so much on the discovery aspect, but more on the conformance and enhancement techniques that exist in the process mining area. For instance, one related to conformance checking is the alignments. So just to recap, let’s say that we want to align the trace given above with the process model below. Well, of course, we can put in a token. And the first activity is easy. A aligns with A. But then you have a choice.
Which of the three paths will I take that best aligns with the given trace? And note that none of the three paths perfectly aligns. Therefore we have an algorithm that does this for us, and always returns the optimal alignment. You cannot do this locally. You need to do this for the whole run through the process model. And I’ll show you why. So let’s say that we already decide to take activity B, because that’s the next matching activity. Then if we continue, the full alignment is shown here. So you need to do a move on log on C, and later move on model on C. And then E and F are mismatching. And this has cost four.
So four times you cannot move synchronously in the trace and the model. Another option that we could have taken is to take this C. And at first, it looks the same as the other C. And of course, when we align this, we again have a move on model, and then a move on log on C, and a move on model and move on log on E and F. The cost for this alignment is also four. Of course, there’s also the third option that, at first sight, doesn’t seem to be the best one. But in the end, we only need to repair C. And B, D, E, and G are all fitting. And this alignment has only a cost of two.
There are only two misalignment steps. So this is how an alignment works. And by calculating an alignment, you relate each event in the event log to an activity in the process model, or you explicitly note that there’s no match at that point in time. So alignments are also crucial not only to see how well the data and the model match, but also to extend the model later on. And we’ve shown you in ProM how you can do this on real data. So we’ve calculated alignments, and this was the projection that you got. You know how frequent an activity is executed, but you also see that one activity was skipped a couple of times.
And once you have this, you can also project timing information. For instance, here, the darker the are color red, the more time is spent in or before that activity.
In this week, we also discussed several ways to do social network analysis in ProM. For instance, we can discover several social networks that relate and show you how several users work together within a process, or within a case. But we’ve also shown you that you can use the dotted chart for this analysis. And even process discovery algorithms can be changed so that they show you a social network.
We have also shown you several process mining case studies. So using real data, this is what other people found on a particular data set. And in the article provided, we also provide many more links to several case studies for you to explore and to learn from. Much more than I can ever present you in these lectures.
We’ve also shown you a refined process mining framework that goes in a bit more detail than describing that there’s discovery conformance and enhancement techniques. This relates different types of input data, different types of process models, with the several activities that you can do within process mining. And note that this is not a comprehensive framework. Not all process mining activities are included, but it gives a very good idea of what is possible in process mining. Then, in the previous lecture, we’ve positioned each of the activities of the refined process mining framework on the spaghetti-lasagna continuum. So certain activities you can do on spaghetti type of processes. But for others, you need more lasagna type of processes.
You need to have a close relation between your process model and event data to have a high predictive value, for instance.
So this is all that I have to teach you in this MOOC. In the next week, week four, there are no video lectures. So no new content. But we have enough time for you to practice. So there’s a peer assignment, where you are asked on real data to apply everything that you’ve learned so far. And you will write this down in a report style. And your peers, your fellow learners, are asked to evaluate your assignment, and you’re asked to evaluate the assignment of others. Next to this, we will also have a final test that summarizes everything that you’ve learned more from the theoretical side. And of course, we invite you to continue the discussions.
So now that you know the basics of process mining, where do you see applications, where do you see challenges? I’m looking forward to discussions and, of course, your peer assignments and final test grades. And from this end of the video, I would like to wish you all the best. And I hope to see you again, and maybe even in real life. Thank you for watching and following this course.