AI Model Training Center
End-to-end pipeline for the goods ETA model · current production: v4.2.1
Training Pipeline
Step 1 · Data Cleaning
1,842,037 rows · 0.4% outliers removed
Step 2 · Missing Value Handling
KNN imputation on partial GPS/FOIS fields
Step 3 · Feature Engineering
47 features incl. corridor density, SR impact
Step 4 · Model Training
Gradient Boosted Trees · 5-fold CV
Step 5 · Validation
Holdout 15% · stratified by division
Step 6 · Accuracy Evaluation
MAE 6.2 min · RMSE 9.8 min
Step 7 · Model Deployment
v4.2.1 promoted to production 24-Jun-2026
Current Model · v4.2.1
Records Used
1,842,037
Features
47
MAE
6.2 min
RMSE
9.8 min
Accuracy
91.4%
Status
Deployed
Feature Importance (top)
Training History
| Version | Date | Records | MAE | RMSE | Accuracy | Status |
|---|---|---|---|---|---|---|
| v4.2.1 | 24-Jun-2026 | 1,842,037 | 6.2 | 9.8 | 91.4% | Deployed |
| v4.2.0 | 17-Jun-2026 | 1,798,221 | 6.9 | 10.4 | 90.6% | Archived |
| v4.1.5 | 03-Jun-2026 | 1,750,140 | 7.4 | 11.2 | 89.8% | Archived |
| v4.1.0 | 20-May-2026 | 1,701,892 | 8.1 | 12 | 88.5% | Archived |