Please enable JavaScript.
Coggle requires JavaScript to display documents.
Business Process Variants - Coggle Diagram
Business Process Variants
Process variants
A process variant is a group of traces that share similar characteristics while being different from other variants. Any trace that belongs to a process variant cannot be a part of other variants at the same time. Since no two different variants are alike, the union of all process variants constitutes the whole event log.
Example
Divide the log into three variants based on attribute COUNTRY.
Divide the log into variants having traces whose events, executions order, and lengths are equal.
Divide the log into variants based on a trace clustering result.
Relevant questions
Why do the executions on a particular variant takes a longer time than the average to complete?
What activities are often skipped in a variant but never or rarely skipped in another variant?
Perspective
Control-flow
comparing variants in terms of the occurrence of events and their executions orders within traces.
Performance
comparing variants according to performance characteristics or performance measures.
Resource
comparing variants from the perspective of resource that work on a case and across cases.
Data
identify differences in data objects or data attributes manipulated by different variants, comparing decision logic across multiple variants.
Example
Highlight the differences between variants whose activity is skipped and variants whose activity never or rarely skipped depending on the data objects or data attributes manipulated throughout variants.
Approach
Process Mining
Replay & Alignment
Replay an event log on top of a process model and quantify the discrepancies between them.
Identify deviations and uses it to characterize process variants.
Machine Learning
Use machine learning techniques to extract features that characterize process variants.
Explainable
make the model (that produces the extracted features that characterize process variants) explainable i.e., the set of rules are easily explained in human terms.
Techniques
Considered dimensions
Process Perspective (control-flow, performance)
Control flow
Resource
Data
Performance (cycle time, waiting time between activities, elapsed time)
Multiple perspective
Outcome (rule-based, model-based, descriptive)
Rule
Descriptive statistics
Alignment matrix
Annotated process model (PN, DG, BPMN, C-BPMN, TS)
Process Model (PN, DG, BPMN, C-BPMN, TS)
Ensemble classifier
Rhythm-eye view
Matrix based representation of Differential Graph
Family of Algorithms (process mining, machine learning)
Hidden Markov Model
Formal Concept Analysis
Fuzzy Mining
Association Rule
Alignment analysis
Causal Relation Analysis
Contrast Itemset Mining
Decision Tree
Frequent Pattern Mining
Clustering
Difference Model Analysis
Log Replay
Difference Model Analysis (TGraph)
Transition System Mining
C-BPMN mining
Prime event structures
Partial synchronized products
Ensemble learning via stacking
Rule mining
Monitoring and configuring rhythm eye visualization
Process map mining
Perspective and differential graph
Evaluation of Data
Real life
Artificial
Input
Event log
Process Model
Both
Implementation
ProM/RapidProM
Apromore
Others
Unification of analysis method
Hybrid
Extract discriminant patterns and visualizing them
Process model based
Compute and visualize frequent sequence pattern on normative process model
Compute and visualize alignments across different variants
Generative
Visualizing differences among variants graphically
(i.e., using a process model)
Process model based
Comparing process model against each other
Comparing process model against normative process model
Discriminative
Extract discriminant patterns among variants
Process model based
Alignments
Token Replay
Indicate frequent path on normative process model
Vector based
Tandem & maximal repeats
n-gram
Challenge
Data and resource-aware variants analysis
See process variants -> perspective -> data & resource
Avoiding insignificant correlations
Use statistical confidence to detect differences and return only significant correlations
Avoid spurious correlations
Infer causation rather than correlation e.g., from "customers who are dissatisfied have longer cycle time" to "long cycle time implies customer dissatisfaction".
Actionable variant analysis
Insights should be actionable i.e., lead to a process change that effectively improve the performance of business process
Reinforcement learning
Each agent maximizes its cumulative reward to generate corresponding process executions.
Move the agent to another process variants so that the new performance shows how far the process variants are from each other.