dhruvchourey/Anomaly_model

Anomaly-Detection Model — manual tester (v3)

Gradio UI to hand-test the trained flood / sensor-anomaly model.

Enter one set of sensor readings + a node, get the calibrated anomaly %. The reading is appended to a 25-hour calm baseline so the model's rolling / rate features have context, then scored through the real serving pipeline (anomaly_service.AnomalyScorer).

Scope: good for flood / out-of-range checks. A stuck sensor cannot be reproduced from a single reading. On this dataset the anomaly % behaves as a 2-band indicator (~13 % normal, ~96 % anomalous).

Files

file	role
app.py	Gradio interface
anomaly_service.py	serving wrapper (AnomalyScorer)
anomaly_features.py	feature engineering + rule layer
anomaly_model_v3.joblib	trained bundle (HGB + calibrator + rule spec)
baseline_history.csv	25 calm hours per node, for feature context