To prevent spam users, you can only post on this forum after registration, which is by invitation. If you want to post on the forum, please send me a mail (h DOT m DOT w DOT verbeek AT tue DOT nl) and I'll send you an invitation in return for an account.

Interpreting the Heuristic Miner Fitness Value

aronlindbergaronlindberg Posts: 8
edited September 2016 in - Usage
I am running PRoM 6.6 using the package "HeuristicsMiner". When I run my model it displays a fitness value in the lower lefthand corner of the visualization. However, I don't know how to interpret these values, and the various academic publications I've found don't seem to shed any light on this.

Hence, my question is: How do I interpret the fitness value that comes out of the HeuristicsMiner?
Post edited by aronlindberg on

Comments

  • @JBuijs: I've been watching your MOOC at FutureLearn, which covers the heuristics miner. However, it doesn't seem to cover this particular issue. Do you know how to interpret the fitness statistic?
  • Dear Aron, I looked through the Heuristic Miner code and found that apparently the following class is responsible for calculating the displayed fitness value:

    org.processmining.plugins.heuristicsnet.miner.heuristics.miner.fitness.ImprovedContinuousSemantics

    The JavaDoc says:
    /**
    * Calculates the fitness of <code>HeuristicsNet</code> objects
    * in a population based on the ratio of activities
    * in a log that could be parsed (or replayed) without
    * problems. The ratio contains a punishment component based
    * on the amount of problems encountered during the log replay
    * <i>and</i> the amount of traces with parsing problems. In a
    * nutshell, this punishment factor benefits the
    * <code>HeuristicsNet</code> objects that have fewer problems
    * scattered in fewer traces.
    * <p> Note that the parsing semantics of this fitness measure
    * is a continuous one (i.e., the log replay does not stop
    * when problems are encountered).
    *
    * @author Ana Karla Alves de Medeiros
    *
    */

    I guess the academic publication corresponding to this is this one: http://is.ieis.tue.nl/staff/aweijters/WP166.pdf
    The CPM measure seems to be matching with the description in the JavaDocs.

    I hope this helps.
  • @fmannhardt -- thanks--this is somewhat helpful. It seems that I am encountering some negative values here. I don't quite see how this can be possible--do you see that this could be possible?
  • Sorry, I did not have the time to look into this. Maybe the PhD thesis of Seppe vanden Broucke may help you:
    http://seppe.net/seppe/Seppe vanden Broucke PhD.pdf
    He wrote a whole section about shortcomings of the heuristic miner.

    Btw:
    Soon, there might be a new heuristic miner with some new features. Stay tuned :smile:
Sign In or Register to comment.