Water Tank Consumption Random Forest

This model predicts hourly household water consumption in liters.

Model

  • Algorithm: RandomForestRegressor
  • Number of trees: 200
  • Maximum depth: 12
  • Minimum samples per leaf: 5

Input features

  • hour
  • day_of_week
  • is_weekend
  • lag_1_hour
  • lag_2_hours
  • lag_24_hours
  • lag_168_hours
  • average_previous_3_hours
  • average_previous_24_hours

Output

Predicted consumption in liters/hour.

Training data

The source dataset is licensed under CC BY 4.0.

Evaluation

Method One-hour test MAE
Seven-day baseline 11.622 L/hour
Linear Regression 11.947 L/hour
Random Forest 10.981 L/hour

24-hour forecasting

Method Daily MAE
Seven-day baseline 43.00 L/day
Random Forest 42.95 L/day

Random Forest improved one-hour predictions, but the two methods were effectively tied for recursive 24-hour forecasting.

Limitations

  • Trained using only one anonymized Swedish apartment.
  • Not validated using the final ESP32 hardware.
  • Requires 168 hours of historical consumption.
  • May not generalize to other households.
  • Not suitable for confirming water leaks.
  • Leak-detection experiments were not sufficiently reliable.

Files

  • model.skops: trained model
  • metadata.json: features, configuration, and metrics
  • example_input.json: example model input
  • requirements.txt: required package versions

Intended use

Educational smart water-tank forecasting and experimentation.

Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support