Page 283 -

10.2 ProM 265

the attributes of some event. However, either way, the ordering of events is lost. This
illustrates that data mining tools, like the mainstream BI products, are data-centric
rather than process-centric.

10.2 ProM
As discussed, existing BI products are data-centric and limited when it comes to
more advanced forms of analysis. Data mining tools offer more “intelligence”, but
are also data-centric. Systems that are process-centric, e.g., BPM/WFM systems,
simulation tools, and process modeling tools, focus on Play-out rather than Play-in
or Replay (cf. Sect. 1.5). Therefore, the traditional software platforms are not use-
able for process mining. This triggered the development of various stand-alone pro-
cess mining tools. In 2002, there were several, rather simple, process mining tools
available, e.g., MiMo (α-miner based on ExSpect), EMiT (α-miner taking transac-
tional information into account), Little Thumb (predecessor of the heuristic miner),
InWolvE (miner based on stochastic activity graphs), and Process Miner (miner as-
suming structured models) [102]. At this time, several researchers were building
simple prototypes to experiment with process discovery techniques. However, these
tools were based on rather naïve assumptions (simple process models and small but
complete data sets) and provided hardly any support for real-life process mining
projects (scalability, intuitive user interface, etc.). Clearly, it did not make any sense
to build a dedicated process mining tool for every newly conceived process discov-
ery technique. This observation triggered the development of the ProM framework,
a “plug-able” environment for process mining using MXML as input format. The
goal of the first version of this framework was to provide a common basis for all
kinds of process mining techniques, e.g., supporting the loading and filtering of
event logs and the visualization of results. This way people developing new process
discovery algorithms did not have to worry about extracting, converting, and load-
ing event data. Moreover, for standard model types such as Petri nets, EPCs, and
social networks default visualizations were provided by the framework.
In 2004, the first fully functional version of ProM framework (ProM 1.1)was re-
leased. This version contained 29 plug-ins: 6 mining plug-ins (the classic α miner,
the Tshinghua α miner, the genetic miner, the multi-phase miner, the social net-
work miner, and the case data extraction miner), 7 analysis plug-ins (e.g., the LTL
checker), 4 import plug-ins (e.g., plug-ins to load Petri nets and EPCs), 9 export
plug-ins, and 3 conversion plug-ins (e.g., a plug-in to convert EPCs into Petri nets).
Over time more plug-ins were added. For instance, ProM 4.0 (released in 2006) con-
tained already 142 plug-ins. The 27 mining plug-ins of ProM 4.0 included also the
heuristic miner and a region-based miner using Petrify. Moreover, ProM 4.0 con-
tained a first version of the conformance checker described in [80]. ProM 5.2 was
released in 2009. This version contained 286 plug-ins: 47 mining plug-ins, 96 anal-
ysis plug-ins, 22 import plug-ins, 45 export plug-ins, 44 conversion plug-ins, and 32
filter plug-ins. Figure 10.3 shows two plug-ins of ProM 5.2. This version supports
all of the process mining techniques presented in this book. For example, each of the