Inter-Case Features in PPM

Most PPM models treat each case in isolation. Inter-case features capture the surrounding system state — work-in-progress, resource contention, simultaneously running cases — that materially affect prediction targets like remaining time.

Examples of inter-case features

Number of active cases at the prefix cut-off.
Activity-level workload — cases currently at activity a.
Resource workload — cases handled by resource r.
Queue lengths per activity / resource.
System age — time since the log started or since last drift.

Why it matters

Congestion effects make remaining time strongly dependent on current system load; a model unaware of workload underestimates time during busy periods and overestimates during quiet ones.

Seminal treatment: Senderovich et al. 2017 — Intra- and inter-case features for predictive process monitoring (sources/2017-senderovich-intra-inter-case). Demonstrated significant accuracy gains from inter-case features on benchmark logs.

Implementation

Compute workload features per prefix at training and inference time.
Concatenate with intra-case features before encoding or model input.
Can be combined with any model family (classical ML, LSTM, Transformer).

concepts/predictive-process-monitoring · concepts/trace-encoding · concepts/remaining-time-prediction