Inductive miner in ProM

Hi, and welcome to this lecture where we show how the inductive miner can be used in ProM. So the inductive miner is, again, one of the algorithms that is able to take the event log data and produce a process model. As we’ve seen in the previous lecture, the inductive miner is one of the few mining algorithms that actually guarantees sound process models. And the inductive miner works by repeatedly finding a split in the event log between within the trace– how to split it. Then detects the operator that describes the splits, and then continues on the sublogs. We also applied the inductive miner on our example data, and this was the Petri net that was discovered.

It’s sound, it can perfectly replay the data that we see, and precision generalization and simplicity are all OK. So now, let’s switch to ProM and see how we can apply the inductive miner on event logs. So let’s import the running example of the loan process with the partial traces again. And now let’s search for the inductive miner. So when you search, you see three variants. Let’s start with the process tree mining variant. So the output is a process three. And you have several options. So I won’t go into too much detail, but the infrequent option is usually the best one. And I’ll provide some advanced material for those that are interested to know something more.

The noise threshold indicates how much behavior can be ignored by the inductive miner. And as you can read here, if you set this to zero, all behavior will be included. Here you can select the classifier, or what the activity notion is. And here you can click on to find more information about this particular algorithm. When we run this algorithm on this data, this is the process tree that you find. So you see a sequence root operator with register application. And then after that, you do a parallel split of three activities. Then you have a choice between reject and accept. And finally, you send the decision email.

Here on the top, you can switch the visualization of the same process tree. So you can visualize it as BPMN if you like this notation.

And there’s also another process tree visualization that you can choose. However, we can also translate this to a Petri net. So we can press the Play button and then convert it to a Petri net. Now you get the results that you will also get when you would pick the other inductive miner variant that’s already at the first step results in a Petri net. And this is actually the Petri net that we have discussed on the slides. And that’s actually the process that we want to discover. However, there’s also the inductive visual miner. So let’s go back to our event log and again search for inductive miner. And now we see and select the inductive visual miner option.

When you do this, you already immediately get a visualization. And first, it discovers a process model. And when it has found a process model, it automatically aligns the data with the process model. And when it has finished, it will show you an animation. So this is an interactive visualization. On the right, you can reduce the slider of the activities so you get– when you move this down– you get fewer activities. And in the end, you even get zero activities. But for this, we can show all activities. And similarly for the edges, you can remove some edges. Although now, there’s not really a threshold that will remove edges.

You can still change the classifier, but you can also show, for instance, where the data deviates from the process model that is discovered. Or you can show the queue length, the service time, and the sojourn time of the activities. And that will show you where bottlenecks are in the system. For instance, here, when you take sojourn time, you see that check system is taking relatively the longest. And you can use this to analyze where temporal bottlenecks are. On the bottom, you see the intensity of the events, so many events are active. And you can also jump to particular moments in the event log.

And on the bottom right, you see the exact time and date of the point you’re looking at. You can also export the model, and even the animation. So now, let’s look at the real data sets. So let’s remove all data, and let’s import the BPI 2012 challenge event log.

Again, let’s apply the inductive visual miner because it gives nice results.

It takes a while to mine a process model. And you’ll see that it contains many activities, and the procedure is not very clear.

So what we could do is, again, filter this, on only including the A and O activities. So we filter using simple heuristics. This is all fine. This is fine. We select traces that end with any activity. And here, we select all activities that start with A and O. And let’s rename the event log. And now, let’s again start the inductive visual miner and see what the result is. And you see that you get a rather clear process model. So you can zoom in. And in the meantime, it’s aligning the data. And it’s already finished. So when we inspect, we see how often particular activities have been executed.

And also here on this part, you see that 5,000 out of 13,000 take the bottom route, and then go through this path. Here, they split in parallel activities. And also, these are executed in parallel. And you can further analyze this process model in a bit more detail. And as I’ve shown you before, you can also analyze the timing perspective. Maybe the sojourn time here is interesting. And you can see that, for instance, here, there’s a relatively large delay of 21 or 16 days before it’s canceled and created and send in a couple of days. And again, you see these tokens flowing.

So I hope you see that the inductive visual miner is a rather flexible algorithm that allows you to quickly investigate real data. So now you’ve seen how you can apply the inductive miner on event logs. So– and especially the visual inductive miner is really helpful in seeing whether there’s a clear process structure hidden in your data. And furthermore, the visual inductive miner also allows you to replay the data and animate how cases flow through the process model and already have an insight on where time is spent in the process.

In this course, we will discuss one more process discovery algorithm in more detail in the next lecture. And then that concludes all for this week. So I hope to see you in the next lecture.