What is the difference between process controlling and process monitoring?
Process controlling: deals with the analysis of historic process executing.
- Input: event logs that relate to a particular period of time.
- Provides insights into whether the general objectives of a process have been met and whether the KIPs are in line.
- Example: offline activity, which involves logs of completed process executions.
Process monitoring: concerned with the quality of currently running process instances.
- Input: event logs of individual process instances or cases
- It works with objectives and rules that are formulated for these individual cases, and triggers counteractions when these rules are violated
- Example: continuous online activity, which involves events of currently running instances.
What are the five major challenge for log data extraction?
- Correlation challenge: problem of identifying the case an event belongs to.
- Timestapm challenge: Challenge to work properly with timestamps stems from the fact hat many information systems do not consider logging as a primary task.
- Snapshots challenge:
- Refers to the issue of having log data availble for a certain period of time.
- For long-running processes, we might not be able to observe all cases of the log. - Scoping challenge: Scoping of the event spectrum is a challenge when the available information system does not directly produce event logs.
- Granularity challenge: interested in conducting event log analysis on a conceptual level for which we have process models defined.
What are the limitations of the alpha-algorithm?
• Requires behaviorally complete log :
Problematic for processes with, e.g., lots of parallelism
It is desirable to use algorithms that can explicitly distinguish likely and unlikely behavior in order to generalize when logs are not complete.
• Cannot distinguish short loops from parallelism.
All of these models lead to b || c
• Cannot deal with noisy data
Event logs often include cases with a missing head, a missing tail, or a missing intermediate episode. Furthermore, there may be logging errors with events being swapped or recorded twice. Such unlikely behavior should not distort the process discovery result.
What are approaches to address problems of completeness?
- Try to balance four essential quality criteria:
1. Fitness: degree of log behavior that a process model is able to replay. It can be defined based on the fraction of event patterns represented by the model or based on the fraction of cases that can be replayed in the model.
2. Simplicity: the resulting process model should be readily understandable. It can be measured using different complexity metrics for process models such as model size or degree of structuredness.
3. Precision: degree of behavior that is allowed by the model, but not observed in the logs. We can easily create a process model that allows the execution of all tasks in any arbitrary order with potential repetition. However, we can hardly learn any specifics of the process from such a model.
4. Generalization: ability of a process model to abstract from the behavior that is documented in the logs. A discovery technique that is able to generalize helps to work with incomplete behavior.
- Automatic process discovery:
Q: What is the actual process model?
Automatic process discovery is concerned how a process actually works in reality. Event logs can be used as an input to evidence-based process discovery.
Automatic process discovery utilizes event logs for the generation of a corresponding process model. Now the event logs are valuable to find a process model where no model existed before, and to adjust an existing model according to how the process really works.
- Performance analysis
Q: what is the performance of the process?
Process analyses such as flow analysis suffer from the fact that the average cycle time for each task in the process model needs to be estimated.
Also, strong assumptions are required such that the behavior of the process is not influenced by the load.
- Using event logs, we inspect the actual behavior of a process and compare it with insights from process analysis.
- Conformance checking
Q: to which extent are the rules of the process model followed?
Q: to which extent are the rules of the process model followed?
Conformance checking is a collection of techniques that compare a set of event logs with a set of constraints or an existing process model.
- The conformance checking can be utilized in order to determine how often the process is executed as expected and, if not, at which stage deviations can be found.
- Here, event logs help to understand either where the model needs to be corrected or where the behavior of the participants working in the process has to be adapted.
