Ranking all 134 U.S. freight corridors by disruption risk — from the federal Commodity Flow Survey to a live, interactive risk explorer.
The highest-vulnerability corridor is not the most dangerous one to lose. ClearPath separates raw freight concentration from network position — and finds that Dallas–Fort Worth, only #4 by vulnerability score, produces the single worst rerouting outcome in the country.
🔗 Live Dashboard · Figma Project Brief · Full Case Study
Geospatial ML · Network Analysis · Federal Data · Business Analytics Capstone (MISM 6214)
| Metric | Value |
|---|---|
| Random Forest LOOCV accuracy | 90.3% across all 134 areas |
| National freight value analyzed | $18.04T (within 8% of the BTS benchmark) |
| Worst-case rerouting cost | $5.28T — Dallas–Fort Worth failure |
| CFS areas ranked | 134 |
| Corridor pairs modeled | 17,822 directed pairs |
| Raw shipment records ingested | 94,023 |
- The Problem
- Target Audience
- How the Team Framed It — Miro & Figma
- Data Pipeline
- The Centroid Bug
- Vulnerability Scoring
- Machine Learning — Random Forest
- Gravity Model
- Network Simulation — The Counterintuitive Finding
- Rerouting Playbook
- Hidden Risk: Memphis
- Key EDA Findings
- Automated Monitoring
- The Dashboard
- Challenges & Resolutions
- Recommendations
- Tech Stack
- Repository Structure
- How to Run
- Data Sources & References
U.S. domestic freight networks are dangerously fragile. Too much shipment value flows through too few geographic corridors, often relying on a single transportation mode. When one port or corridor fails, the disruption cascades nationally — costing billions and exposing critical industries.
The 2024 Francis Scott Key Bridge collapse made this concrete: a single piece of infrastructure failed, regional freight dropped 25% within one month, and delay-related costs ran at an estimated $12M per week. Baltimore is not even a top-5 freight corridor by value — yet the disruption required a coordinated federal and state response.
| Stat | Value | Meaning |
|---|---|---|
| U.S. logistics costs (2024) | $2.58T | 8.8% of GDP — well above the pre-pandemic 7.4–7.8% range |
| LA–Long Beach freight value | $7.96T | A single corridor; one disruption there affects every industry |
| CFS freight zones mapped | 134 | The geographic resolution the 2022 CFS now enables |
| Shipment records analyzed | 94,023 | Primary CFS API analysis layer |
| Bridge-collapse freight drop | 25% | Regional, within one month |
| CFS data vintage | 2022 | Newest available subarea data |
The gap we fill: existing studies use simulation or global data. No publicly available tool had used the 2022 CFS subarea data to map domestic freight vulnerability at the corridor level with ML. ClearPath Analytics is that tool.
ClearPath is built for three decision-makers who currently lack a ranked, geographically explicit vulnerability index:
| Audience | Who | What they need |
|---|---|---|
| Shipment Managers | C.H. Robinson, J.B. Hunt, XPO, Flexport, UPS, DHL | Real-time rerouting recommendations when a primary corridor is disrupted — which alternative, what added cost/ton, how many extra miles |
| Policymakers & Infrastructure Planners | MARAD, FHWA, Port of LA/Long Beach, Port of Houston, FMC | Evidence-based investment priority rankings — which corridors to fund, and the dollar cost of inaction (FY2026 PIDP grants: $488.6M) |
| High-Exposure Corporations | Pharma (Pfizer, J&J), electronics (Apple, Dell), autos (GM, Ford), retail (Walmart, Target) | Commodity-specific corridor risk scores — "Is my pharma supply chain exposed if LA–Long Beach fails? What's the nearest alternative?" |
Before any data was touched, the team mapped the entire causal chain — from disruption to economic consequence — so the analysis stayed anchored to actionable outputs rather than interesting-but-unusable findings.
START → estimate freight value per area → freight flows through the 134 CFS areas, then a fork: a risk-cascade lane (concentration → allocation strain → reduced capacity → shipment delays → inventory shortages → rising logistics costs → supply-chain vulnerability) and an analytics lane (2022 CFS data → aggregate → normalize value/tonnage → score → rank → flag high-risk commodities). Both converge on "which areas require action?" → logistics teams reroute, planners build capacity, policymakers fund improvements.
The five-part brief: Business Problem → Target Audience → Steps We Are Taking → Predicted Solution & Outcome → Final Product.
All primary data originated from two federal agencies: the U.S. Census Bureau's 2022 Commodity Flow Survey (CFS) API and the Bureau of Transportation Statistics (BTS) TIGER/Line shapefile repository. The four CFS datasets were retrieved programmatically via paginated API calls (94,023 rows), validated against published CFS cross-tabulation totals.
⚠️ Security note: an early version of the extraction script had a Census API key hardcoded in source. It was caught during code review, regenerated, and replaced with aCENSUS_API_KEYenvironment variable before the repo was made public.
Census CFS API (94,023 rows)
→ Centroids (INTPTLAT/INTPTLON from shapefile)
→ Distance Matrix (17,822 Haversine corridor pairs)
→ Port Proximity (NTAD intermodal facilities)
→ Features Master (21-column ML-ready table)
→ Vulnerability Scores (134 CFS areas ranked)
| Dataset | Source | Rows | Description |
|---|---|---|---|
cfs_freight_area.csv |
Census CFS API | 94,023 | Freight value + tonnage by CFS area and commodity code |
cfs_exports.csv |
Census CFS API | 102 | Export flows by origin state and commodity |
cfs_hazmat.csv |
Census CFS API | 48 | Hazmat freight — high-risk corridors needing specialized routing |
cfs_temp_controlled.csv |
Census CFS API | 308 | Cold-chain freight — food/pharma vulnerability |
centroids.csv |
BTS TIGER/Line shapefile | 134 | Census INTPTLAT/INTPTLON per CFS area (post-bug-fix) |
distance_matrix.csv |
Team-computed | 17,822 | Haversine distances for all 134×133 directed corridor pairs |
features_master.csv |
Team-engineered | 134 | 21-column ML-ready feature table; zero nulls |
vulnerability_scores.csv |
Team-computed | 134 | Composite vulnerability index for all 134 areas |
During gravity-model construction, a significant upstream error surfaced. The original centroids.csv had been geocoded by area name rather than by shapefile coordinates — collapsing 66 of 134 areas onto incorrect coordinates:
- Buffalo, Rochester, Albany → all placed on New York City's lat/lon
- Baltimore, MD → placed in Columbia, SC
- San Diego, Fresno, Sacramento → all on a single generic California point
This produced 118 zero-distance corridors and silently wrong distances on 13,266 of 17,822 corridors (74%).
The fix: read INTPTLAT/INTPTLON directly from 2022_CFS_Areas.shp (joined on GEO_ID = CFS22_GE_1) — a guaranteed-inside, authoritative Census coordinate — and recompute all distances with Haversine. Zero zero-distance corridors remain. All four derived geo files were rebuilt and backed up as *.orig.csv.
The Spearman rank correlation between original and corrected vulnerability scores was ρ = 0.978 — the geographic story held, but absolute distances and rerouting costs were only valid after the fix.
A second data-quality issue was caught separately: the original scores summed freight values across all three SCTG commodity hierarchy levels (2-, 3-, and 4-digit), triple-counting most shipments and inflating the national total to $112T. The corrected pipeline uses only the
COMM=0aggregate row per area → $18.04T, within 8% of the BTS-published benchmark of $19.6T.
Each of the 134 CFS areas receives a composite score:
vuln = (val_norm × 0.60) + (ton_norm × 0.40)
Shipment value carries 60% weight because high-value freight (pharmaceuticals, electronics, motor vehicles) generates disproportionate economic disruption per unit of volume. Tonnage carries 40% to capture bulk-freight corridors where infrastructure stress is highest even at lower per-unit value.
Robustness: the index was run at 60/40, 50/50, and 40/60 weightings — all yield Spearman ρ ≥ 0.99, and 9 of the top 10 corridors are identical across all weightings. The rankings are robust to the weighting choice.
| Rank | CFS Area | Freight Value | Vuln. Score | Tier |
|---|---|---|---|---|
| 1 | Los Angeles–Long Beach, CA | $7,963.8B | 0.882 | High |
| 2 | Houston–The Woodlands, TX | $4,234.6B | 0.719 | High |
| 3 | Chicago–Naperville, IL-IN-WI | $4,290.3B | 0.521 | High |
| 4 | Dallas–Fort Worth, TX-OK | $3,757.1B | 0.500 | High |
| 5 | Remainder of Texas | $2,166.3B | 0.434 | High |
| 7 | Remainder of Pennsylvania | $2,254.8B | 0.345 | High |
| 14 | Seattle–Tacoma, WA | $2,123.7B | 0.243 | High |
| 70 | Memphis–Forrest City, TN-MS-AR | $862.5B | 0.085 | Low |
Tiers (percentile-based): High (top 10%), Medium (75th–90th), Low (below 75th) → 14 High, 20 Medium, 100 Low. Gini coefficient for freight value = 0.461 (concentration is real but not extreme); 66 areas are needed to cover 80% of national freight value.
With only 134 rows, a standard 80/20 split would leave ~27-sample test sets — statistically unreliable. Instead, Leave-One-Out Cross-Validation (LOOCV) is used: each area serves as its own test case exactly once. The classifier uses class_weight='balanced' to compensate for the 100:20:14 Low:Medium:High imbalance.
Result: 90.3% LOOCV accuracy.
| Feature | Importance |
|---|---|
| Tonnage | 0.326 |
| Value | 0.285 |
| Num Commodities | 0.187 |
| Val/Ton Ratio | 0.076 |
| Seaport Distance | 0.065 |
| Neighbor Distance | 0.042 |
| Is Metro | 0.019 |
| Dominant Commodity | 0.000 |
Dominant commodity type contributes exactly zero predictive importance. It's how much freight moves through a corridor — not what kind — that determines its vulnerability tier. This directly contradicts the intuition that pharma or electronics corridors are inherently more vulnerable than bulk-commodity corridors.
| Tier | Precision | Recall | F1 | Support |
|---|---|---|---|---|
| High | 0.80 | 0.86 | 0.83 | 14 |
| Medium | 0.73 | 0.55 | 0.63 | 20 |
| Low | 0.94 | 0.98 | 0.96 | 100 |
| Overall | 0.90 | 0.90 | 0.90 | 134 |
The lower Medium-tier recall (0.55) reflects genuine ambiguity at the middle tier. Two High-tier areas (Remainder of Kentucky, Seattle–Tacoma) were predicted Medium; their scores (0.2452, 0.2427) sit just above the 90th-percentile cutoff — genuine borderline cases, not misclassifications.
A spatial gravity model scores all 17,822 directed corridor pairs:
gravity_ij = (VAL_i × VAL_j) / distance_ij²
High-gravity corridors move the most freight value over the shortest distance, so a disruption there propagates the largest economic shock. The gravity_norm (0–1) column serves directly as the edge weight in the NetworkX graph.
The Philadelphia–Reading–Camden corridor emerged as the highest-gravity nationally — specifically between the PA and NJ state parts, separated by just 16.96 miles. The extremely short distance combined with high freight value at both endpoints produced a gravity score nearly twice that of the next-ranked corridor (New York–Newark NJ ↔ NY, 60 mi). This correctly surfaces dense, short-haul intra-metro flows as the highest-impact disruption scenarios.
Script:
scripts/gravity_model.py· Output:supply_chain_data/03_outputs/corridor_gravity_scores.csv(17,822 ranked rows)
A K=8 nearest-neighbor NetworkX graph was built (134 areas as nodes, Haversine distance as edge weights). Each of the 14 High-tier areas was simulated as a failure (node removal), then Dijkstra shortest-path found every origin-destination pair whose route was affected. Rerouting cost:
cost ($) = extra_miles × (node_TON × 1,000 short tons) × $0.08/ton-mile (BTS standard truck rate)
The most important finding: Dallas–Fort Worth ranks only #4 by vulnerability score but produces the single worst rerouting outcome — because of its structural network centrality. LA–Long Beach is #1 by score but only 4th-worst by rerouting cost, because the K=8 topology offers multiple coastal bypass routes.
| CFS Area | Vuln. Rank | Rerouting Cost | OD Pairs | Extra Miles |
|---|---|---|---|---|
| Dallas–Fort Worth, TX-OK | #4 | $5,284.78B | 484 | 32,954 |
| Remainder of Pennsylvania | #7 | $3,977.46B | 1,028 | 30,695 |
| Remainder of Illinois | #6 | $3,282.03B | 426 | 19,144 |
| Los Angeles–Long Beach | #1 | $1,487.95B | 109 | 7,151 |
| Chicago–Naperville | #3 | $1,279.99B | 314 | 8,759 |
| Houston–The Woodlands | #2 | — (resilient) | 0 | — |
| Seattle–Tacoma | #14 | — (resilient) | 0 | — |
Houston scores #2 by vulnerability but produces zero rerouting cost — the K=8 network absorbs its failure entirely. A lower-connectivity K=4 topology would surface Houston's fragility. The network topology assumption is itself a design choice that shapes the conclusions — documented as a known limitation and Phase 2 extension.
Script:
To be submitted/09_network_rerouting.py· Output:failure_simulation_results.csv,To be submitted/05_rerouting_detail.csv
Alternative corridors pre-identified for the five highest-cost failure scenarios:
| If This Corridor Fails | Primary Reroute Path | Avg. Added Miles | Affected OD Pairs |
|---|---|---|---|
| Dallas–Fort Worth, TX-OK | Remainder of Texas → Remainder of Oklahoma | 68.1 mi | 484 |
| Remainder of Pennsylvania | Philadelphia-Reading-Camden → Pittsburgh-New Castle-Weirton | 29.9 mi | 1,028 |
| Remainder of Illinois | Chicago-Naperville → Remainder of Indiana | 44.9 mi | 426 |
| Los Angeles–Long Beach, CA | San Jose-SF-Oakland → San Diego-Chula Vista-Carlsbad | 65.6 mi | 109 |
| Chicago–Naperville, IL-IN-WI | Remainder of Illinois → Remainder of Wisconsin | 27.9 mi | 314 |
For areas embedded in dense metro clusters, average added distance per shipment is under 45 miles — nearby alternatives absorb rerouted volume without large detours. For areas that are the primary node in a larger region (DFW, LA–Long Beach), it jumps to 65–68 miles — there's no equally positioned alternative nearby.
11. Hidden Risk: Memphis
Memphis–Forrest City ranks #70 of 134 by overall vulnerability score (0.0847) — firmly Low tier. A policymaker scanning the ranked table would have no reason to flag it. Yet commodity-level EDA revealed it handles 4.5% of national pharmaceutical freight value — third nationally, behind only LA–Long Beach (11.5%) and Chicago (6.3%) — because the FedEx SuperHub routes overnight pharmaceutical distribution through Memphis.
This is a mismatch between aggregate vulnerability (Low) and commodity-specific concentration risk (Top 3). A general-purpose ranking built on total VAL and TON will always miss this. Because the risk is driven by a single carrier's hub-and-spoke architecture rather than geography, the correct mitigation is carrier-specific: maintain a secondary overnight carrier (UPS, DHL) for time-sensitive pharmaceutical shipments.
This illustrates why commodity-specific concentration analysis at the SCTG-4 grain — layered on top of geographic scores — surfaces hidden risks aggregate rankings miss. A systematic SCTG-4 scan across all 134 areas is the natural Phase 2 extension.
- Freight concentration is broader than the LA–Long Beach headline. Gini = 0.461; top 5 areas hold only 17.5% of national value; 66 areas needed to cover 80%. → Target the top 30–40 corridors, not just the top 5 ports.
- Bug corrected; vulnerability ranking preserved. SCTG triple-count inflated the national total by 3.42×. After correction: $18.04T, within 8% of the BTS benchmark. Spearman ρ = 0.978.
- Census suppression affects 45.91% of commodity rows.
VAL=0is intentional Census suppression of small/confidential cells, not missing data. All commodity analysis uses SCTG-2 as canonical grain; SCTG-4 reserved for the Memphis case study. - Pharmaceutical concentration centers on Memphis, not the Northeast. Top 5: LA–Long Beach (11.5%), Chicago (6.3%), Memphis (4.5%), DFW (4.2%), Boston (3.9%). Memphis is #3 because of the FedEx SuperHub — not manufacturing.
- Remainder-of-State zones are a distinct bulk-freight risk class. Non-metro Remainder zones carry 42.9% of national tonnage despite only 29.1% of value. Four of the top 10 vulnerability areas are Remainder zones — a structurally different risk profile requiring separate policy response.
digest_runner.py monitors BTS freight indicators for the top 15 CFS areas and sends email alerts when any area's weekly volume drops ≥15% below its 4-week rolling average. It runs serverless on GitHub Actions via a weekly cron schedule — no infrastructure required.
Weekly cron (Mon 12:00 UTC) → pull live BTS indicators → compare vs 4-week rolling avg → email alert if drop ≥15% → log to digest_log.csv
In its first live run against real BTS data, it correctly flagged a 74.1% week-over-week drop in the New York–Newark corridor — exactly the early signal that turns a 72-hour reactive scramble into a 4-hour execution of a pre-approved plan.
Workflow:
.github/workflows/weekly_digest.yml— also supportsworkflow_dispatchwith a--demomode (injects a synthetic disruption). Email via SendGrid (SENDGRID_API_KEY/FROM_EMAILrepo secrets).
🔗 Open the live Streamlit dashboard →
Built with Python · Streamlit · GeoPandas · Plotly · NetworkX.
Vulnerability score by area & risk-tier distribution — all 134 CFS areas ranked, with the High / Medium / Low split.
K=8 nearest-neighbour graph, each area linked to its eight neighbours and colored by risk tier.
Node-removal run across all 14 High-tier areas, totalling ~$17T in rerouting cost.
The headline chart: vulnerability score vs rerouting cost — network position beats raw score.
Origin-destination paths that reroute when Dallas–Fort Worth, the worst-case node, fails.
Top 100 corridors drawn on the map, line thickness ∝ gravity score.
Per-area drilldown — risk tier, freight value, and tonnage for any selected CFS area.
The selected area's eight graph neighbours and its distance to the nearest intermodal port.
422 SCTG categories — flagging $2.1T of hazmat and $3T of temperature-controlled freight.
| Challenge | Impact | Resolution |
|---|---|---|
| Centroid geocoding bug | 74% of corridor distances corrupted (13,266 / 17,822) | Read INTPTLAT/INTPTLON from shapefile via GEO_ID join; rebuilt all derived geo files |
| SCTG triple-counting | National total inflated to $112T (actual $18.04T) | Use only the COMM=0 aggregate row per area; validated vs BTS benchmark |
| API key in source code | Census key hardcoded & committed | Key regenerated; replaced with CENSUS_API_KEY env var before going public |
| Census suppression (45.91% of rows) | Commodity drilldowns unreliable below SCTG-2 | SCTG-2 as canonical grain; SCTG-4 only for Memphis (high volume, low suppression) |
| 134-row dataset | 80/20 split → ~27-sample test sets, unreliable | Leave-One-Out Cross-Validation |
| K=8 topology masks fragility | Houston & Seattle show zero rerouting cost | Documented as a limitation; K=4 analysis flagged as Phase 2 |
Shipment Managers — Pre-negotiate backup carrier capacity on Dallas–Fort Worth first (worst simulated outcome despite only #4 vulnerability rank). Treat Memphis as a named pharmaceutical contingency. Subscribe to the weekly digest for 72-hour early warning.
Policymakers — Prioritize the 14 High-tier areas for investment review — DFW, Remainder of Pennsylvania, and Remainder of Illinois first. Broaden the lens beyond the 5 largest port cities. Treat rural Remainder-of-State zones as a separate investment category.
Implementation — No new infrastructure required to begin. The weekly digest is live on GitHub Actions; the rerouting playbook and investment brief are generated directly from corridor_gravity_scores.csv, risk_tier_output.csv, and failure_simulation_results.csv.
Python · Pandas · GeoPandas · NetworkX · scikit-learn · Random Forest · Haversine · Census CFS API · BTS TIGER/Line · Dijkstra Shortest Path · GitHub Actions · SendGrid · Plotly · Streamlit · Figma FigJam · Miro · LOOCV · Gravity Model · Spearman Correlation
ClearPath-Analytics/
├── .github/workflows/
│ └── weekly_digest.yml # serverless weekly freight-drop digest (cron + manual)
├── assets/screenshots/ # dashboard screens, Miro map, Figma brief
├── scripts/
│ ├── README.md # gravity-model pipeline + centroid-bug writeup
│ ├── rebuild_centroids.py # 1. fix coordinates + distance matrix
│ ├── rebuild_features.py # 2. fix geo-derived feature columns
│ └── gravity_model.py # 3. score + rank all 17,822 corridors
├── supply_chain_data/
│ ├── 01_api_data/ # 4 raw CFS API pulls (freight, exports, hazmat, temp)
│ ├── 02_shapefiles/ # TIGER/Line county + 2022 CFS areas shapefiles
│ ├── 03_outputs/ # gravity scores, vulnerability scores, charts
│ ├── 04_references/ # literature + bibliography
│ ├── 05_features/ # centroids, distance matrix, features master (+ .orig backups)
│ ├── Supply_Chain_Data_Dictionary.docx
│ └── data_inventory.txt # provenance + centroid-bug correction log
├── To be submitted/ # final deliverable bundle (corrected CSVs, scripts, RF notebook, charts)
│ ├── 01_vulnerability_scores_corrected.csv
│ ├── 02_risk_tier_output.csv
│ ├── 03_corridor_gravity_scores.csv
│ ├── 04_failure_simulation_results.csv
│ ├── 05_rerouting_detail.csv
│ ├── 06–08_*.py # rebuild + gravity scripts
│ ├── 09_network_rerouting.py # K=8 graph + Dijkstra failure simulation
│ ├── 10_Risk_Tier_Classifier_RF.ipynb
│ └── 11_digest_runner.py
├── digest_runner.py # weekly monitoring job
├── failure_simulation_results.csv
├── vulnerability_scores.csv
└── subscribers.csv # digest recipients
pip install pyshp # plus pandas, numpy, matplotlib
python scripts/rebuild_centroids.py # 1. fix coordinates + distance matrix
python scripts/rebuild_features.py # 2. fix geo-derived feature columns
python scripts/gravity_model.py # 3. score + rank corridorsEach script is idempotent and backs up any file it overwrites to *.orig.csv (only on first run, so backups always hold the original pre-fix data).
pip install pandas numpy
export SENDGRID_API_KEY=... # optional — omit to dry-run
export FROM_EMAIL=...
python digest_runner.py # live mode
python digest_runner.py --demo # inject a synthetic disruption for demosSee To be submitted/09_network_rerouting.py (K=8 graph + Dijkstra) and To be submitted/10_Risk_Tier_Classifier_RF.ipynb (LOOCV Random Forest).
- U.S. Census Bureau — 2022 Commodity Flow Survey (CFS) API — freight value & tonnage by CFS area and SCTG commodity code
- Bureau of Transportation Statistics — TIGER/Line 2022 — county boundaries + 2022 CFS Areas shapefile (134 freight zones)
- NTAD — intermodal port/terminal facility locations
- BTS — $0.08/ton-mile standard truck freight rate; national freight benchmark ($19.6T)
- Literature on intermodal disruption resilience, geographic concentration risk, and the 2025 State of Logistics — see
supply_chain_data/04_references/bibliography.txt
Built as the MISM 6214 Business Analytics capstone. Dashboard screenshots and the Miro/Figma artifacts are captured from the live project environment. Code and corrected datasets are in this repository.