asats/bengaluru-ac-predictor

Bengaluru AC Purchase Predictor

This model predicts whether you should purchase an air conditioner for the next 2 months in Bengaluru, Karnataka, India.

How it works

The model uses current month's weather summary features to classify if the upcoming 2 months will be uncomfortably hot.

Key Assumptions

1. Bengaluru climate: Tropical savanna, ~920m elevation, moderate year-round temperatures
2. Peak summer: March-May (highest AC need)
3. Monsoon: June-September (rain reduces AC need)
4. Winter: December-February (no AC needed)
5. AC needed when:
   • Average max temperature > 31°C in next 2 months
   • Heat index > 32°C

   • 15 days above 30°C across next 2 months

   • Dry heat: low rainfall + temp > 29.5°C

Training Data

• 7 South Indian cities: Bengaluru, Chennai, Hyderabad, Kochi, Mysuru, Coimbatore, Mangaluru
• Time period: 2010-2024
• Source: Open-Meteo historical weather API
• Features: Temperature, humidity, precipitation, heat index, apparent temperature

Model

• Type: Gradient Boosting Classifier
• Features: 13 monthly weather summary features
• Target: Binary (0 = No AC needed, 1 = Buy AC)

Usage

from inference import ACPredictor

predictor = ACPredictor("ac_model.pkl")

# Current month summary (example: March data)
current_month = {
    'month_tmax_mean': 33.5,
    'month_tmax_max': 36.2,
    'month_tmin': 22.1,
    'month_tmean': 27.8,
    'month_rh': 58.0,
    'month_precip': 15.0,
    'month_hi': 34.2,
    'month_hi_max': 38.5,
    'month_apparent_max': 35.0,
    'month_apparent_mean': 30.1,
    'month_days_above_30': 18,
    'month_days_above_32': 8,
    'month_days_hi_above_32': 20,
    'month': 3
}

result = predictor.predict(current_month)
print(result)
# {'ac_needed': 1, 'confidence': 0.92, 'reasoning': 'High avg max temp: 33.5C; High heat index: 34.2C; Low rainfall: 15.0mm'}

Results

• Training samples: 895
• Model accuracy: See metadata.json for full metrics

Author

Created by asats via Hugging Face Agent.