dinsdag 27 september 2016

analyse road traffic fine

Below I analysed the Road Traffic Fine-process for the Futurelearn-course 'Process Mining with Prom' 

First the summary:
The proces has 150370 process instances with 561470 events.
Always the same start event (start fine) but not always the same end event. (actually there are 7 different end events)

Now we look at the process summary, where we can see that there is data from januari 2000 till june 2013. There are 11 event classes. Some cases contains 20 events (10 event classes) some ony 2.
The mean event per case is 4.

In the explore event log we can see that almost 38% of the log has the events
Create-Send-Insert-Add penalty-Send. Almost 31% of the cases has only the events Cre-Pay.

Looking at the Event dotted chart we see a rather complex chart. Some filtering is needed to make it more easy.

Here a dotted chart were we can see the relation between the event-names and the index in the trace.
You can see that create is always on the first place and so on.

After filtering the log using simpe heuristics we can create a chart on timestamp and conceptname.
Sorting on timestamp of the first event result in the following chart.
You can see the process is not very constant and fluctuated a bit. Duration is longer since about 2005.

Analysing the process

First using the alfa miner (based on the ordering relations) to geta basic understanding of the process

With the heuristic miner (based on the number of events) we see a model and the number of events.

Below the Inductive miner

And inductive visual miner, with activities set to 0,55.

Here the fuzzy mining models looks as follows

Here with a closer focus

Fot conformance checking and further analysis I used the inductve miner. Seems the most clear to me.

Creating the model

As we see in the explore ent log almost 38% of the instance has the events Create-Send-Insert-Add penalty-Send and almost 31%  has only the events Cre-Pay. Here we can see this in a model.
Here I filtered the log using simple heuristic

Confirmance checking

Using the inductive miner and the log I first create the folowing model. (onformance analysis)

move-log-fitness is 0.99 Fitness is OK but this model is rather complex. 
So, below I created a more simpler model 

move-log ftness of this model is lower (= 0.82) but the model is more simple and easy to understand

Also tried the replay a log on petri net for performance/conformance checking
Average throughput time = 4,27 months
Max throughput time= 69,53 months
Observed period = 163,9 months

In the element statistics we can see that the event Payments has to longest throughput, with an avg. of 3,26 months

Social analysis

Social anaysis on this proces is  bit hard (for me). Here I tried the work together option.
Hope to find some better social analysis later.

It's clear to me that although I have a basic understanding of the mining algoritmes now, also experience is needed to make a good process mining analysis. It's aslo clear that (in good hands) Prom can be a very powerful tool.

Geen opmerkingen:

Een reactie posten