Project 10
An ensemble anomaly detection system that flags unusual water quality measurements across Calgary's Bow and Elbow river networks, covering 82 parameters at 7 monitoring sites
Streamlit application
Four pages covering site-level anomaly maps, flagged readings, time-series drill-downs, and configurable alert thresholds.
Map
Monitoring site locations with anomaly counts and severity indicators across the watershed
Anomalies
Flagged readings from the ensemble detector with parameter-level breakdowns and vote counts
Time series
Parameter trends over time with anomaly overlays and rolling statistics per monitoring site
Alerts
Configurable alert thresholds and notification rules for operational water quality monitoring
Key results
Ensemble precision
0.75
Majority-vote ensemble combining four detection methods
Ensemble F1
0.74
Balanced precision-recall from Isolation Forest, LOF, 3-sigma, and z-score
Parameters
82
Water quality parameters monitored across 7 river sites
Methodology
Fetched watershed water quality data from Calgary Open Data. Pivoted long-form measurements into wide-form, engineered rolling statistics, rate-of-change, and z-score features per site. Applied Isolation Forest, Local Outlier Factor, 3-sigma rule, and z-score detection, then combined all four into a majority-vote ensemble for balanced precision and recall.