K-Means, Isolation Forest, and Random Forest were used to organize and prioritize unusual readings.
PROJECT 02 / Anomaly Detection · Environment
Water Quality Analysis
A project that searched water measurements for unusual patterns so potential pollution, sensor problems, or emerging environmental risks could be investigated sooner.
DATASETMultivariate water quality measurements
SIZERecord count not preserved in public documentation
SOURCECoursework dataset, public attribution pending
MEASURED EVIDENCE
What can be verified.
Metrics and outputs drawn from the project artifacts—not estimates added for presentation.
The workflow separates potential environmental anomalies from readings that may reflect sensor or data quality issues.
The result prioritizes records for expert review rather than declaring water safe or unsafe.
VALIDATION
Scope based evidence only: the project demonstrates a reproducible anomaly screening workflow. Dataset size and a labeled performance benchmark were not documented, so no accuracy claim is presented.
Why this project matters
Water quality datasets contain many measurements at once, and a dangerous or unusual reading can be difficult to spot in a large table. This project grew from the need to separate routine variation from patterns that may signal pollution, changing environmental conditions, or faulty sensors.
What was developed
The analysis cleaned and explored the measurements, grouped similar observations with K-Means, and used Isolation Forest and Random Forest methods to surface unusual readings. The goal was to turn a dense environmental dataset into a prioritized set of patterns worth investigating.
What it means
The analysis does not declare that water is safe or unsafe. It creates an early warning view that helps analysts focus attention on unusual locations or measurements and decide where testing, field inspection, or expert review may be needed.
04 / WHERE THIS WORK APPLIES
From project to practical use.
This approach can be used by utilities, environmental agencies, farms, watershed programs, aquaculture operations, manufacturers, and sensor networks that monitor drinking water, rivers, lakes, groundwater, or industrial discharge.
05 / LIMITATION
What the project does not solve yet.
Anomaly labels and geographic coverage are limited, so some unusual readings may reflect sensor error or local conditions rather than true environmental risk.
06 / NEXT ITERATION
Future advancement
Integrate geospatial and time series context, validate anomalies with domain experts, and support streaming sensor data with real time alerts.
TECHNICAL TOOLKIT
PandasSciPyK-MeansIsolation ForestRandom Forest