Leontjeva et al. 2015 — Complex Symbolic Sequence Encodings

Foundational paper on trace encodings for outcome prediction. Systematically compares encoding schemes on real logs — establishes the taxonomy of encodings that subsequent PPM surveys adopt.

Contribution

Frames traces as complex symbolic sequences — preserving both activity labels and per-event data payload.
Compares encodings:
- Boolean (activity occurrence)
- Frequency (activity counts)
- Simple index-based (first / last / sliding-window)
- Complex index-based (one attribute block per position)
- HMM-based
Outcome = binary classification (positive vs negative business outcome).

Significance

Sets the encoding taxonomy later cited by Teinemaa et al.'s outcome-PPM benchmark (sources/2016-teinemaa-outcome-ppm-review) and the Verenich survey (sources/2019-verenich-survey-ppm). Establishes that data-aware encodings beat control-flow-only encodings — a claim echoed by all subsequent neural PPM work.

Connections

Concepts: concepts/trace-encoding · concepts/outcome-prediction · concepts/predictive-process-monitoring Authors: entities/chiara-di-francescomarino · entities/marlon-dumas · entities/fabrizio-maggi