Definition
Accuracy of the Airport Collaborative Decision Making (A-CDM) system's predicted turnaround completion time at the moment downstream operational decisions (gate reassignment, pushback sequencing, de-icing dispatch) are committed: measuring prediction fidelity at the point where errors translate into operational cost.
The Measurement Gap
Airport operations decisions (gate reassignment, pushback sequence, de-icing truck dispatch, air traffic clearance requests) are made based on predicted turnaround completion times. The operational cost of these decisions depends entirely on the quality of the prediction that justified them. A premature pushback sequence blocks other aircraft; a delayed de-icing dispatch causes a de-icing queue; a gate reassignment made too early creates unnecessary passenger confusion.
Traditional airport performance metrics measure on-time departure rates and taxi time: they measure outcomes, not prediction quality. A-CDM systems issue multiple turnaround predictions as milestones are achieved, but no standard metric measures which of those predictions were accurate at the point when an irreversible downstream decision was committed based on them.
Decision supported:
Are the turnaround predictions that drive gate, pushback sequencing, and de-icing dispatch accurate enough to justify the decisions being made based on them, or is the hub's operational rhythm being disrupted by acting on predictions that regularly turn out to be wrong? TPA provides the measurement signal that determines whether the A-CDM system's predictions warrant operational decisions at the confidence currently being placed in them. Low TPA at a hub station during peak periods indicates that predictions should carry wider uncertainty buffers before triggering downstream actions, or that real-time milestone data feeds need to be added to the prediction model.
Formula
Plain language: For each turnaround where the A-CDM system issued a predicted completion time that triggered a downstream operational decision, compare the prediction to the actual off-block time. Normalize the error by turnaround duration to prevent short turns from dominating. Average across all measurement-period turnarounds. Subtract from 1 to express as an accuracy score.
Notation:
TPA = 1 − (1/N) × Σ |Predicted_completion_i − Actual_completion_i| / Turnaround_duration_i
Variables:
| Variable | Definition | Source | Unit |
|---|---|---|---|
| Predicted_completion_i | A-CDM target off-block time (TOBT) or estimated off-block time at the decision-commit point for turnaround i | A-CDM platform / AODB | Timestamp |
| Actual_completion_i | Actual off-block time (AOBT) for turnaround i | A-CDM platform / AODB / ACARS | Timestamp |
| Turnaround_duration_i | Scheduled turnaround duration for normalization | Schedule system | Minutes |
| N | Number of turnarounds in the measurement period | n/a | Count |
Also compute directional bias separately: are predictions systematically optimistic (predicted completion earlier than actual (leading to premature pushback sequencing) or pessimistic (later than actual) leading to missed departure slots and unnecessary gate holds)? Directional bias matters more than average accuracy for operational planning.
Proposed formula, pending validation pilots.