TransX Reefer Risk Predictor

TransX Reefer Analytics

Fleet-wide analytics platform for reefer shutdown risk prediction, cost modeling, maintenance optimization, and repair pattern analysis.

Quick Stats

Fleet Units --

Avg Fleet Risk --

Shutdown Rate --

Avg Cost/Engine-Hr --

PM Overdue --

Unit Selection

Reefer Unit

Trailer Number

Trip Details

Route

Season

Service Type

Leg Miles

Temperature (set)

Maintenance History

Auto-filled from unit data

Days Since PM

Late PM Count

PM Compliance %

Repair Cost ($)

Repair Count

Unique Repairs

Select a unit and trip details,
then click Calculate

Training Data Context

Training base rate3.34%

Training trips (2025)23,473

Shutdowns in training785

Model AUC0.84

Features used83

Risk Tiers

LOW — <2.0% shutdown probability

MODERATE — 2.0–4.7%

HIGH — 4.7–12.9%

CRITICAL — >12.9%

Filters

Location

Lane

Service Type

Model Year

Tier

Make

Search Unit

Units

Avg Risk

Critical

High

Shutdowns

SD Rate

Risk Tier Distribution

fleet baseline below

Group By

-- | Sort:

Click a row to add it as a filter. Bars show tier mix; orange/red tints flag elevated avg risk vs the current slice.

Repair & Maintenance Pattern (current slice vs fleet baseline)

per-unit averages

Metric	Slice Avg	Fleet Avg	Δ vs Fleet	Distribution

PM Delay vs Risk & Shutdown Rate (current slice)

PM Status	Units	Avg Risk %	Shutdown Rate %

Fleet Risk Ranking

click column header to sort

#	Unit	Trailer	Model · Yr	Make	Last Lane	Svc	Risk %	Tier	Trips	SDs · Rate	90d Reps	Comp Codes	Last PM	PM Overdue	Total Repair $

Loading model comparison…

Make × Model × Year × Season

Risk is re-scored for each unit under each of the 4 seasons (other features held constant). Highlights which cohorts are most weather-sensitive.

Make

Model

Year

Season Highlight

Fleet Seasonal Baseline (avg risk if every unit ran in that season)

Cohort × Season Heatmap

Sort by:

Make	Model	Year	Units	Winter	Spring	Summer	Fall	Swing max−min

Cells are color-scaled by risk magnitude. Highlight a season above to outline that column. Swing column shows how weather-sensitive the cohort is.

Most Weather-Sensitive Cohorts (largest gap between worst and best season)

Cohort	Units	Worst Season	Worst Risk	Best Season	Best Risk	Swing

Reefer Model Deep-Dive

Per-model history: shutdown rates, age cohorts, component patterns, and action items. Uses observed historical trip data.

Worst Performing Cohorts

Model + year combinations ranked by shutdown rate — identifies specific vintages to target

Model	Year	Age	Units	SD Rate %	Avg Risk %	Cost/Trip	Elevated+	Top Component Issue

Component Vulnerability by Model

Avg repairs per unit — darker = more repairs. Identifies which components fail on which models.

Risk by Model × Year

Full breakdown showing age effect per model

Model	Year	Age	Units	SD Rate %	Avg Risk %	Cost/Trip

Actionable Recommendations by Model

Loading repair pattern analysis...

Baseline Failure Rate

All trips

With Repair in 14d

Without Repair in 14d

Diag-Only (No Fix)

Time window:

Component Group Failure Rates

Trips with this component group repaired in the pre-trip window — sorted by failure rate. Lift = how many times riskier vs baseline.

Component Group	Trips	Shutdowns	Failure Rate	Lift	Risk Bar

Repair-to-Trip Turnaround

How soon after the last repair the unit was dispatched — shorter gaps = higher risk

Diagnosis Resolution

Was the diagnosis followed by an actual component fix before the trip?

Failure Rate by Repair Reason

Was the pre-trip shop visit a PM, general repair, warranty, etc.?

Reason	Trips	Failure Rate	Lift

Description Keywords

Keywords in repair descriptions (14d window) and their associated failure rates

Keyword	Trips	Failure Rate	Lift

Dangerous Repair Combinations

Two component groups repaired within 14 days before a trip — combinations with highest failure rates

Component A	Component B	Trips	Shutdowns	Failure Rate	Lift	Risk Bar

Repair Sequences (Order Matters)

Last two work orders before a trip, in chronological order. The sequence reveals whether issues escalated or went unresolved.

First Repair		Second Repair	Trips	Shutdowns	Failure Rate	Lift

Repeat Repair Impact

Same component group repaired 2+ times in 30 days before a trip vs single occurrence — repeat = unresolved issue

Component Group	Pattern	Trips	Shutdowns	Failure Rate	Lift

Detailed Comp Code Breakdown

Individual repair codes with 5+ occurrences in the pre-trip window

Code	Group	Trips	Shutdowns	Failure Rate	Lift

Loading cost per hour...

Fleet Avg Total $/Hr

PM + Repair + Fuel

Avg PM $/Hr

Avg GENREP $/Hr

Reefers Analyzed

Select Reefer

Fuel Price ($/L)

Make

Model

Year

PM Cost

$/hr

GENREP Cost

$/hr

Fuel Cost

Grand Total

$/hr

PM Cost Breakdown by Code

Each PM track predicted independently, costs summed

Loading CPM data...

Avg GENREP CPM

Avg PM CPM

Avg Total CPM

Fleet Units

Select Unit

Pred GENREP

Pred PM

Pred Total

Total Miles

Rail Miles

Highway Miles

Year

Age

Type

Reefer

Dominant Shop

# Shops

SHAP Cost Drivers

Per-unit SHAP values in CPM units. Positive = increases cost, negative = reduces cost.

GENREP Drivers

PM Drivers

What-If Analysis

Adjust feature inputs and get predicted GENREP CPM, PM CPM, and Total CPM instantly.

Use selected Unit Lookup values as baseline

total_miles

rail_share

pm_orders_per_1k_mi

genrep_orders_per_1k_mi

reactive_pm_share

pm_delay_mean_days

pm_delay_pos_rate

pm_delay_neg_rate

pm_order_count

consignee

Field definitions

total_miles: planned/expected total miles for this scenario.

rail_share: fraction of miles on rail. Example 0.90 = 90% rail, 10% highway.

pm_orders_per_1k_mi: PM order frequency per 1,000 miles.

genrep_orders_per_1k_mi: GENREP order frequency per 1,000 miles.

reactive_pm_share: share of PM done in mixed (repair+PM) context, 0 to 1.

pm_delay_mean_days: average PM timing delay in days (positive = late, negative = early).

pm_delay_pos_rate: fraction of PM cycles that are late beyond threshold.

pm_delay_neg_rate: fraction of PM cycles that are early beyond threshold.

pm_order_count: expected PM order count for the period/profile.

consignee: customer/consignee profile context.

Total Units

High CPM Units

Low CPM Units

Avg Pred Total CPM

CPM Ranking (Lowest to Highest)

Fleet-wide CPM ranking with HIGH/MEDIUM/LOW bands (P20/P80 thresholds)

Outlier note: extreme CPM units can inflate error metrics. Retained for transparency.

#	Unit	Pred CPM	Pred Cost	Total Miles	Rail Miles	Hwy Miles	Band

Global SHAP importance shows average CPM contribution magnitude by feature across the fleet. Higher mean |SHAP| = stronger average influence on CPM prediction.

GENREP Driver Pattern

PM Driver Pattern

Feature Glossary

Feature	Description

3-Month Pilot Validation

Capture fleet-risk snapshots, compare predictions vs observed outcomes, and track intervention impact on fill-rate over the pilot window.

Pilot Start

Pilot End

Window --

Snapshot Library

Snapshots persisted in webapp/data/snapshots/ are loaded from the repo. Click "Save Current Snapshot" to download a JSON you can commit there.

Import JSON

Captured	Label	Source	Units	Avg Risk	Shutdowns	Tier Mix	Notes	Actions

Compare Snapshots

Baseline: → Current:

Δ Units

Δ Avg Risk

New Shutdowns

Hit Rate (HIGH+)

Tier Upgrades

Tier Downgrades

Unit	Group	Baseline Risk	Current Risk	Δ Risk	Tier (Base → Now)	New SDs	90d Reps Δ	Repair $ Δ

Pilot Groups

Assign units to Intervention (receiving the proactive action) or Control. Stored in your browser; export to JSON to share with the team.

Import Config

Intervention 0

Avg Risk-- Shutdowns-- SD Rate-- 90d Reps Avg--

Control 0

Avg Risk-- Shutdowns-- SD Rate-- 90d Reps Avg--

Unit	Tier	Risk %	Last Lane	Shutdowns	90d Reps	Fillable	Assignment

Data Refresh & Model Retrain

Stage updated source files for the next model refresh. This page validates that all required files are present and flags filename / extension mismatches before the project team runs the Python retraining pipeline.

Required Files

Uploaded

Validated

Needs Review

Required Source Files

Files staged here are kept in your browser session only. The project team runs python scripts/refresh.py server-side to retrain.

File	Target Path	Source	Uploaded Name	Size	Uploaded	Status	Action

Optional Files (pipeline continues without them)

File	Location	Impact if Missing
Quarterly Telematics Q1–Q4	`data/telematics/quarterly/Daily Telematics Update -YTD 2025 Q*.xlsx`	No alarm / restart / low-fuel features
Fuel Data (per year)	`data/fuel data/MMA 20XX Fuel Data.xlsx`	No fuel-fill features
Weather Cache	`data/training/weather_cache.csv`	Auto-fetched from Open-Meteo API

Pipeline Steps

What scripts/refresh.py does once the project team runs it. Typical runtime: ~10–15 minutes.

1
Validate data files exist — confirms all required files are at their expected paths.
2
Feature engineering (build_trip_features_v3.py) — reads raw data, builds 199-column trip-level feature CSV → data/training/trip_features_v3.csv.
3
Model training (build_risk_engine.py) — calibrated XGBoost on temporal train/test split (Jan–Sep / Oct–Dec) → data/risk_engine/*.
4
Webapp export (export_webapp.py) — converts model artifacts to webapp-compatible JSON → webapp/data/*.
5
Repair pattern analysis (repair_pattern_analysis.py) → webapp/data/repair_patterns.json.
6
Validation & smoke test — model loads and scores a sample trip; AUC ≥ 0.70 expected.

TransX Risk Predictor

TransX Reefer Analytics

Trip Risk Predictor

Fleet Dashboard

Model Comparison

Repair Patterns

Cost Per Hour

Cost Per Mile

Quick Stats

Unit Selection

Trip Details

Maintenance History

Training Data Context

Risk Tiers

Filters

Risk Tier Distribution

Group By

Repair & Maintenance Pattern (current slice vs fleet baseline)

PM Delay vs Risk & Shutdown Rate (current slice)

Fleet Risk Ranking

Make × Model × Year × Season

Fleet Seasonal Baseline (avg risk if every unit ran in that season)

Cohort × Season Heatmap

Most Weather-Sensitive Cohorts (largest gap between worst and best season)

Reefer Model Deep-Dive

Worst Performing Cohorts

Component Vulnerability by Model

Risk by Model × Year

Actionable Recommendations by Model

Component Group Failure Rates

Repair-to-Trip Turnaround

Diagnosis Resolution

Failure Rate by Repair Reason

Description Keywords

Dangerous Repair Combinations

Repair Sequences (Order Matters)

Repeat Repair Impact

Detailed Comp Code Breakdown

Select Reefer

PM Cost Breakdown by Code

Scenario Planner (New Reefer)

Asset Details

Maintenance Plan

Expected Usage

Select Unit

SHAP Cost Drivers

What-If Analysis

CPM Ranking (Lowest to Highest)

GENREP Features

PM Features

Validation Checks

GENREP Driver Pattern

PM Driver Pattern

Feature Glossary

3-Month Pilot Validation

Snapshot Library

Compare Snapshots

Pilot Groups

Data Refresh & Model Retrain

Required Source Files

Optional Files (pipeline continues without them)

Pipeline Steps