Introducing the Stigmergic Miner, a temporal extension of the Fuzzy Miner based on computational stigmergy

Workflow mining is a data mining research field for discovering graph models from event logs. Particularly, Fuzzy Mining is a well-known approach, suitable for noisy and unstructured logs. It includes in the model the fuzzy notion of significant/correlated edges and nodes, thus enabling the formation of granular graphs. This paper introduces the Stigmergic Miner, a novel workflow miner that incorporates the concept of time, as an extension of the Fuzzy Miner. It takes into consideration the temporal dynamics that exist within event occurrences. The proposed algorithm employs Computational Stigmergy, a bio-inspired paradigm employed in multi-agent systems. Stigmergy creates a representational space where each sample is transformed as a mark in the scalar space, whose intensity is evaporating over a discretized time. Overlapping marks create persistent stigmergic trails, generating scalar-temporal clustering of samples. With respect to the simply scalar fuzzy clusters, stigmergic clusters are then inherently temporal. Thanks to parametric optimization, different agents can be specialized to detect different temporal patterns, by computing similarity between stigmergic trail and stigmergic archetypes. Since each pattern is represented with a different color, the resulting graph is colored, to highlight different temporal patterns associated with event time series for different branches and nodes of the graph. To evaluate the effectiveness of this approach, early experiments with real-world data are reported, allowing for empirical validation of the proposed miner. An open-source web application has been developed and publicly released.