A new research paper has proposed a standardized challenge framework dubbed AVOCADO to evaluate process mining algorithms specifically for event streams.
Streaming process mining deals with the real-time analysis of streaming data, with event streams requiring algorithms capable of processing data incrementally, according to the authors.
AVOCADO is designed to systematically address the complexities of this domain by providing clear structural divisions that separate the concept and instantiation layers of challenges in streaming process mining for algorithm evaluation.
The initiative seeks to foster innovation and community-driven discussions to advance the field of streaming process mining.
What is the AVOCADO framework for process mining?
The AVOCADO framework evaluates algorithms on streaming-specific metrics like accuracy, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Processing Latency, and robustness.
It provides a standardized framework for assessing algorithms on their ability to process continuous event data while balancing accuracy and performance under drifting real-time conditions.
By using synthetic event data generated from complex process models, AVOCADO offers clear, objectified evaluation criteria that reflect the challenges inherent in working with event streams, the authors wrote. This advances the state of process mining by driving the development of algorithms that can learn the expected behavior from event streams, while meeting the system and resource constraints typical of real-time, large-scale environments.
Ultimately, AVOCADO aims to foster process mining techniques allowing organizations to gain insights into their business processes in real-time, improving decision-making and operational efficiency.
Register for All Access: Process Intelligence 2026!
Addressing process mining challenges
Traditionally, process mining focuses on process discovery – the extraction of process models from event data. These event logs typically represent finite sets of recorded activities, often used to model the flow of a business process.
While event logs provide a detailed record of past events, they are finite and often limited in the scope they capture, which can restrict the ability of process mining techniques to generalize.
In recent years, the need to extend process mining techniques to handle event streams has grown and thus to efficiently handle continuous, potentially infinite sequences of events. Event streams better reflect real-time, dynamic business processes and offer a richer, more comprehensive data source for analysis.
The shift from event logs to event streams in process mining introduces several challenges. Event streams are continuous and real time, requiring algorithms to process data as it arrives. This real-time processing demands algorithms that can incrementally update process models without access to the entire dataset at once, placing constraints on both memory and computational resources.
To address these challenges, AVOCADO is designed to access algorithms on their ability to process continuous event data while balancing accuracy and performance under drifting real-time conditions, according to the authors.
“By using synthetic event data generated from complex process models, AVOCADO offers clear, objectified evaluation criteria that reflect the challenges inherent in working with event streams. This challenge advances the state of process mining by driving the development of algorithms that can learn the expected behavior from event streams, while meeting the system and resource constraints typical of real-time, large-scale environments.”