Prompt

Build anomaly detection system for data streams.

<data_stream>
- Type: {metrics/logs/transactions}
- Volume: {records per second}
- Dimensions: {number of features}
- Normal patterns: {describe typical behavior}
</data_stream>

<anomaly_types>
{point, contextual, collective anomalies}
</anomaly_types>

<business_requirements>
- False positive tolerance: {acceptable rate}
- Latency: {detection speed needs}
- Explainability: {why flagged}
</business_requirements>

Build system:
1. Method selection
   - Statistical methods
   - ML approaches (Isolation Forest, etc)
   - Deep learning (Autoencoders)
   - Ensemble strategies
   
2. Feature engineering
   - Time-based features
   - Statistical aggregates
   - Seasonal adjustments
   
3. Threshold optimization
   - Precision-recall trade-off
   - Dynamic thresholds
   - Anomaly scoring
   
4. Alerting design
   - Alert prioritization
   - Grouping strategies
   - Notification rules

Include evaluation metrics.