Key Performance Indicators consist of performance information derived directly from the event log or by replaying an event log onto a process map. They represent the basis for performance analysis. These performance metrics can be grouped into three classes: process related KPIs, node related KPIs and edge related KPIs.
For the computation of KPIs we take as input an event log and a process map.
The process map can have any number of hierarchical levels and the event log can be at any level of abstraction. Let L be the given event log, and PM be the process map.
3.5.1 Process KPIs
Process KPIs refer either to performance metrics which are measured at the process level and can be computed from event logs without a process model, or to
perfor-3.5. Key Performance Indicators (KPIs) 61 mance metrics which are computed on the level of traces (i.e., process instances) from event logs and a process model.
• Performance metrics computed from event logs without directly using a process model:
1. Number of traces
The total number of traces (i.e., process instances) in L.
2. Number of unique traces
The total number of unique traces in L represented as process instances, where a process instance is unique if the corresponding sequence of events is unique.
3. Arrival rate of traces
The number of traces that arrive per unit time in L.
• Performance metrics computed from event logs with a process model:
1. Number of tting traces
The tting traces are those in which there are no skipped tasks (or tasks which are encountered in the log but not in the model) and no missing edges. The concepts of skipped task and missing edges are presented in Sections 3.5.2 and 3.5.3.
2. Activities in the model but not in the event logs
A complete set of nodes that belong to PM but do not have a corre-sponding activities in L.
3. Activities in the event log but not in the process model
A complete set of activities that are encountered in L but do not have a corresponding node in PM. For these activities we also provide count information: how many times these activities were encountered in the event log.
3.5.2 Node KPIs
In order to compute these performance metrics, both the event log and a process map are required.
1. Execution time of a node
The execution time is computed for abstract nodes only. Let t(i, j), (i ≤ j), be an instance of a sequence of activities in a trace t mapped to the abstract node n. The execution time of n is de ned as the time spent between the moment the rst activity t(i) occurs and the moment the last activity t(j) occurs. For this KPI, several statistics are calculated:
• The average execution time of all instances of node n in L.
• The minimum execution time of all instances of node n in L.
• The maximum execution time of all instances of node n in L.
• The standard deviation of the execution time of all instances of node n in L.
4. Node initialization frequency
The total number of traces in L that start with an instance of node n.
5. Node termination frequency
The total number of traces in L that end with an instance of node n.
3.5.3 Edge KPIs
Edge KPIs are related to the control ow perspective of a process map. Like in the case of node KPIs, for these performance metrics, both the process map and an event log are necessary.
1. Execution time of an edge
The execution time of edge e = (n,n0) is de ned as the total time spent to route a process control from an instance of the source node n to an instance of the target node n0. It is calculated as the time spent from the moment the last activity t(i) in a sequence that is mapped to n occurs, until the rst activity t(j) in a sequence that is mapped to n0 occurs, with the condition that t(j) receives process control from t(i).
For this KPI, several statistics are calculated:
• The average execution time of all instances of an edge e in L.
• The minimum execution time of all instances of an edge e in L.
• The maximum execution time of all instances of an edge e in L.
• The standard deviation of the execution times of all instances of an edge e.
where an instance of an edge refers to the situation in which the control is passed from the source node to the target node of the edge in L.
2. Number of executions
The number of times edge e is executed, i.e. control is passed from the source node to the target node of the edge.
3. Number of times an edge is missing
If t(i, j) is a sequence of activities that has a corresponding ow path in the process map. The edges corresponding to the ow path that do not have an activity in t(i, j) mapped to the source node or to the target node, are considered to be missing.
A formal de nition of these KPIs is presented in Appendix A. The KPIs de ned in
3.5. Key Performance Indicators (KPIs) 63 this section are generic and can be used for performance analysis with any business process model, such as Heuristic Net [24]. We proposed a method for computing these KPIs given an event log and a hierarchical process model. The method is based on the representation of hierarchical process models as process maps.
of the plugins customized/implemented during this master's thesis. In Section 4.2 we explain the customization of the Pattern Abstractions plugin and in Section 4.3 we explain in detail the implementation of the plugin.
4.1 Plugins Overview
In this section, we provide an overview of all plugins that we customized/implemented.
Existing process mining techniques (Dotted Chart Analysis, Conformance Check-ing, Pattern Abstractions, etc.) available in the ProM framework, have been used for the analysis in Chapter 2. Moreover, the Pattern Abstractions plugin has been customized for Philips, in order to use domain knowledge for giving proper names to abstractions. Furthermore, to tackle the research objective presented in Chapter 1, a new plugin needs to be developed to support the computation of performance met-rics and annotation of a process map with the performance information. A process map and an event log are required as inputs for the new performance analysis plugin.
To abstract for low level event logs and create abstractions we selected the Pattern Abstractions plugin. To visualize the hierarchical process models we selected fuzzy maps [12], [11] as these can describe any processes in an intuitive way. Fuzzy maps are hierarchical process models created using the enhanced fuzzy miner that sup-ports the discovery of process maps [12], [11], [16]. In order to be used as input for performance analysis, the process (fuzzy) map has to be exported using the provided export functionality of the Fuzzy miner plugin.
The exported process map and an event log serve as input. These objects are stored in the Object Pool. Since the most suitable component to retrieve and use objects from the Object Pool is the Plugin component, we implemented the Fuzzy Map Performance Analysis Plugin (FMPA) as a class in the Plugins compo-nent. This plugin produces the KPI values and the Annotated Process Map (APM) as output. The performance metrics are useful as long as they can be visualized intuitively, therefore we implemented a visualization class as part of the Visualizer
64