Smart Work Zone Intelligence · SRF Consulting Group

Current Conditions

Average Speed –mph

Volume vs Typical Peak –% of peak

Operating State –

LIVE SPEED (LAST 5 HRS)

✓ Loading current conditions…

Current State

Corridor segments colored by nearest-sensor speed reading

Device Uptime: –

Active Sensors: –

Latest Reading: – ✓

Predictive Risk

Breakdown Probability — ML Pilot

Model Output Horizon 15 Minutes

Breakdown Probability –%

Confidence Level –

PREDICTIVE SUMMARY

Breakdown Probability
15 Minute Horizon: –%

SPEED FORECAST

🕑Projected Average–

Current Speed–

15 MINUTE TREND GRAPH

-30 minNext 15 min

Risk Tier Indicator — Confidence Level: –%

Feature State

Corridor segments colored by the current corridor-wide risk tier (the breakdown model outputs one probability for the whole work zone, not per-segment — every segment reflects the same tier)

FORECAST BREAKDOWN TREND -30 min Next 15 min –%

Moderate Elevated

Model Transparency

Feature Importance — Breakdown Classifier

Which sensor signals drive the model's breakdown probability output (all 21 features)

Note on Jerk features: Upstream Jerk and QD1 Jerk (Shojaat et al. 2026) rank near the bottom because this pilot used synthetic sensor data with smooth Gaussian noise — speed.diff().diff() of smooth curves yields near-random second differences uncorrelated with breakdown events. With real field sensor data, sharp jerk pulses at shockwave arrival would rank these features near the top, providing several additional minutes of advance warning.

Calibration Curve — Probability Accuracy

Gray dashed = perfectly calibrated. Blue = this model. Points on the diagonal mean when the model says 70%, 70% of those events actually happen. Marker size reflects how many holdout points support that bin.

PERFORMANCE & ACCOUNTABILITY Zone Performance Score – –

Overall Score: –

Loading…

Zone Performance –

Contractor Score –

Zone Performance

Monitoring System Outcomes – –

Average Work Zone Speed vs Target

–

Service Reliability Index (Planning Time Index)

–

Incident Impact Duration (avg time to speed recovery)

–

Weekly Breakdown Count–

Avg Event Duration–

Stability Score–

Trend (first half vs. second half of week)–

Contractor Performance –

Lane closure compliance, device uptime, and DMS notification response — last 7 days

LANE CLOSURE COMPLIANCE — LAST 7 DAYS

Device Uptime — Target – –

Notification Response Time — Target under – –

Planned vs Actual Closure Duration Variance –

How the System Works

01

Ingest & Parse

Six field sensors push one observation per minute via the WZDx 4.1 federal standard — speed, volume, occupancy, and device health.

02

Feature Engineering

21 predictive features are computed: rolling averages (5/15 min), rate-of-change, V/C ratio, multi-zone occupancy, and jerk (Shojaat et al. 2026).

03

Train on 80%

Three Random Forest models are trained on the first 80% of observations — chronological split to prevent data leakage and simulate live deployment.

04

Predict & Alert

Every minute, the model predicts speed, queue length, and breakdown probability 15 minutes ahead. DMS messages are triggered automatically.