Monitoring¶
Continuous monitoring is essential to ensure that models remain accurate over time as production data distributions change.
In this stage, we compare the features of incoming production request traffic against our training baseline to calculate feature drift and monitor operational and performance metrics.
Below is the latest generated model performance and feature drift report:
# Model Monitoring Report
**Report Generated At:** `2026-06-06 16:44:33`
**Target Model:** `nyc_taxi_demand_forecaster` (Version: `3`, Alias: `@champion`)
**Monitored Samples:** `71253`
---
## 1. Quality Gates (Inference vs. Target)
The table below compares the performance of the active champion model on
incoming production data against the configured quality gates.
| Metric | Target Gate | Current Value | Status |
| :--- | :--- | :--- | :--- |
| **MAE** | `<= 50.0` | `12.0741` | ✅ PASS |
| **RMSE** | `<= 75.0` | `24.4486` | ✅ PASS |
| **R² Score** | `N/A` | `0.8880` | `N/A` |
---
## 2. Feature Drift Analysis
We monitor for distribution changes in features to identify if model updates
are needed due to changes in consumer patterns.
| Feature | Train Mean (Baseline) | Production Mean | Drift (Diff %) | Drift Status |
| :--- | :---: | :---: | :---: | :---: |
| **pickup_count** | `42.9883` | `48.0728` | `+11.83%` | `🚨 Drift Detected (Warning)` |
| **hour** | `11.90` | `11.90` | `+0.00%` | `Normal` |
| **day_of_week** | `2.96` | `2.96` | `+0.00%` | `Normal` |
| **is_weekend** | `0.2894` | `0.2894` | `+0.00%` | `Normal` |
| **month** | `1.01` | `1.01` | `+0.00%` | `Normal` |
---
## 3. Summary & Actions Required
- **Data Drift**: The main volume feature `pickup_count` shows a `+11.83%` shift from baseline.
- **Model Health**: Model predictions remain within the evaluation thresholds.
- **Recommended Action**: Initiate retraining run because significant drift is detected.