weijuinlee/coastal-risk-ai

coastal-risk-ai predicts whether a flood alert will occur in the next 60 minutes for a location using rainfall, wind speed, tide level, and location metadata.

Model Summary

• Task: binary classification
• Target: flood_event_next_60m
• Forecast horizon: 60 minutes
• Model type: lightgbm

Risk Output

• Probability: 0.0 to 1.0
• Risk labels:
  • low for probability < 0.35
  • medium for probability >= 0.35 and < 0.7
  • high for probability >= 0.7

Evaluation

Validation:

• PR-AUC: 0.5348903399631979
• ROC-AUC: 0.6350447465016593
• F1: 0.613998613998614

Test:

• PR-AUC: 0.5335708698674165
• ROC-AUC: 0.6392812506799013
• F1: 0.600754716981132
• Precision: 0.5264550264550265
• Recall: 0.6994727592267135

Expected Input Schema

• location_id
• rainfall_mm
• wind_speed
• tide_level_m
• flood_prone_flag
• hour_of_day
• day_of_week optional
• month optional

Training Data

This artifact was trained from the normalized coastal risk schema used in the repository:

• timestamp
• location_id
• latitude
• longitude
• rainfall_mm
• wind_speed
• tide_level_m
• flood_alert
• flood_prone_flag

The bundled sample dataset is synthetic and intended for pipeline validation and demo use.

Files

• model.joblib: trained model artifact
• features.json: ordered model feature list
• metrics.json: train and evaluation metrics
• config.json: pipeline configuration used for training
• sample_inference_single.json: example inference payload
• sample_inference_batch.csv: example batch inference payload