I refined my airport surge‑price collusion detection system to use a 2‑minute synchrony detection window, a rolling price‑vs‑demand baseline, and real‑time DBSCAN clustering on (synchrony score, price delta). When a cluster meets the confidence threshold I trigger either a soft‑cap freeze or dynamic dampening, then reward honest drivers with a fair‑play rebate and audit nightly with an isolation‑forest detector.
I use a synchrony detection window of 2 minutes to flag clusters of drivers that go offline then online together.
I compute a synchrony score from a time‑series of driver‑state events per airport.
I maintain a rolling baseline of price‑vs‑demand curves for each hour of the day and each airport.
I apply DBSCAN clustering with ε = 0.2 and min‑samples = 5 on (synchrony score, price delta).
I trigger mitigation when a cluster size ≥ 10 AND synchrony score > 0.8.
I implement a soft‑cap freeze for 5‑10 minutes, setting the surge multiplier to 1.0.
I use dynamic dampening: new multiplier = current multiplier × (1 − 0.05 × (cluster size/10)).
I run a nightly Isolation Forest audit with contamination = 0.01 to validate false‑positive rates.
I publish a weekly price‑integrity report for each airport market.
I observed on 10 Oct 2026 sudden price spikes at busy airports that did not follow normal demand patterns, suggesting possible driver coordination to create artificial supply shocks. I built a system to detect, verify, and mitigate these artificial surges in real time.
Baseline demand‑supply model uses Prophet or seasonal ARIMA.
Driver‑group clustering uses DBSCAN (ε=30 seconds, minPts=5).
Set a hard ceiling on surge multipliers for airport zones (e.g., 2.5×).
Offer small, time‑bounded bonuses for drivers who keep the app active during high‑demand windows ($0.10 per minute).
Penalty engine thresholds: 30 points → 24 h suspension; 10 points → permanent de‑activation.
The system streams lightweight driver‑heartbeat pings every few seconds into InfluxDB, computes a rolling 60‑second count per airport, and flags anomalies when the count drops > 2 σ below the forecast while surge multipliers exceed the cap.