Approximate Inductive Miner for Process Tree Discovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current process discovery algorithms face challenges in efficiently generating sound process models from event logs, particularly when dealing with noise and varying noise levels, often requiring extensive parameter tuning and resulting in high computational intensity and suboptimal models.
Innovation Solution
The Approximate Inductive Miner (AIM) method constructs a process tree by clustering activities in the event log, reducing the search space through filtering and clustering techniques, allowing for automated discovery of sound process models without the need for human parameter input or upfront filtering, thereby handling complex event logs with high precision and low control flow complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If state-of-the-art process discovery algorithms are applied to handle noise in event logs, then the ability to represent event log behavior is improved, but the computational time and complexity increase significantly
Solution Approach 1:
The patent applies preliminary filtering and clustering actions to the event log before the main process discovery algorithm executes. By pre-processing the event log to remove noise and cluster similar activities, the algorithm reduces the input data complexity, thereby decreasing computational time while maintaining precision in representing the event log behavior.
Solution Approach 2:
The patent segments the event log processing into distinct phases: filtering, clustering, and model discovery. This segmentation allows each phase to be optimized independently, with filtering removing noise and clustering grouping similar behaviors, thereby reducing the overall computational burden while preserving precision.
2Measurement precision
If hyper-parameter optimization is performed to determine default parameters for a given event log, then the model quality is improved, but the time required for parameter determination increases
Solution Approach 1:
The patent implements self-service mechanisms where the algorithm automatically determines optimal parameters through clustering analysis of the event log itself, rather than requiring external hyper-parameter optimization. The clustering process inherently reveals the optimal number of clusters and parameter settings, eliminating the need for time-consuming manual or automated parameter tuning while maintaining high model quality.
3Adaptability or versatility
If multiple process models are produced and evaluated to handle varying noise levels, then the adaptability to different event log conditions is improved, but the computational intensity increases
Solution Approach 1:
The patent employs dynamic parameter adjustment based on the characteristics of the input event log. The algorithm automatically adapts its clustering parameters and filtering thresholds according to the noise level and structure detected in the specific event log, eliminating the need to generate and evaluate multiple static models for different noise conditions. This dynamic adaptation maintains versatility while reducing computational intensity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates to an automated discovery of sound process models from process activities of traces of an event log. Provided is a computer-implemented method, the Approximate Inductive Miner (AIM), which exploits clustering techniques to investigate an exponentially growing state space in polynomial runtime. According to an aspect of the invention, the method comprises an integrated parameter suggestion, based on novel cut quality estimates, under consideration of the information loss induced by filtering the event log.