Airline Anomaly Detection Engine
Statistical anomaly detection across five major U.S. carriers using z-score analysis on DOT operational data. Adjust the sigma threshold and filters to surface unusual performance periods.
Detected Anomalies
| Date | Airline | Metric | Value | Mean | Z-Score | Direction | Severity |
|---|
How This Was Built
This dashboard is a fully self-contained HTML application that performs real-time statistical anomaly detection on U.S. airline operational data — no server, no build step, no dependencies beyond a single CDN-hosted charting library.
Data
- Monthly operational metrics for American, Delta, United, Southwest, and Frontier airlines spanning January 2022 through December 2024 (36 months per carrier).
- Sourced from the DOT Bureau of Transportation Statistics T-100 Domestic Segment database and On-Time Performance tables.
- Five metrics tracked: On-Time %, Cancellation Rate, Diversion Rate, Average Delay (minutes), and Total Flights.
Anomaly Detection
- Uses z-score analysis: for each airline + metric combination, the engine computes the rolling mean (μ) and standard deviation (σ) across the selected time window.
- Any data point where
|z| ≥ thresholdis flagged as anomalous. The user-adjustable sigma slider (1.0–3.5) controls sensitivity. - Anomalies are classified by severity: High (≥ 3σ), Medium (2–3σ), or Low (below 2σ).
- All computation runs client-side in plain JavaScript — no Python, no backend, no API calls.
Visualization
- Chart.js 4.4 renders the interactive time-series chart with per-airline trend lines and anomaly markers.
- Anomaly points are drawn as enlarged, highlighted scatter points overlaid on the line series.
- KPI summary cards update dynamically as filters change, showing total anomalies, most-affected carrier, highest z-score, and data coverage.
Architecture
- Single-file HTML — all CSS, JavaScript, and data are embedded inline.
- Dark theme (
#090c14) with responsive sidebar layout that collapses on mobile. - Deployed as a static page on Netlify — zero infrastructure, instant global CDN delivery.
Stack
- HTML5 / CSS3 / JavaScript (ES6+) — no frameworks, no transpilation
- Chart.js 4.4.1 — time-series visualization
- Statistical Methods — z-score anomaly detection, descriptive statistics