Energy-ML: Seasonal Household Energy Consumption Prediction

Description

This project develops machine learning models to predict household electricity consumption in Kerala, India, during monsoon and summer seasons.
The project focuses on understanding energy usage patterns in residential buildings and identifying key drivers like home size, number of occupants, floors, and orientation.

---

Problem Statement

Electricity consumption varies seasonally due to climate and occupant behavior.
This project aims to:

• Predict electricity bills for monsoon and summer months.
• Analyze feature importance to understand energy drivers.
• Provide interpretable insights for energy-efficient home design.

---

Dataset

• Source: Collected as part of the project “Building Self Sustainable Smart Cities through Energy Efficient Homes using Intelligent Design”
• Rows: 500
• Columns:
  • Total Area (sqft) — Total built-up area of the home
  • Number of Occupants — Total residents
  • Number of Floors — Number of floors
  • Orientation — Main direction the house faces
  • KSEB bill in monsoon — Target for monsoon model
  • KSEB bill in summer — Target for summer model

---

Features & Targets

Input Features:

• Total Area (sqft)
• Number of Occupants
• Number of Floors
• Orientation

Target Variables:

• KSEB bill in monsoon
• KSEB bill in summer

---

Methodology

• Preprocessing:
  • One hot encoding for categorical features
  • Standard scaling
• Models:
  • CatBoost Regressor (primary)
• Evaluation:
  • 5-fold cross-validation
  • R² score calculation
  • Feature importance analysis

Separate models were trained for monsoon and summer to capture season-specific patterns.

---

Results

Monsoon Model:

• R² (CV): 0.71
• Top Feature: Total Area (70.9%)

Summer Model:

• R² (CV): 0.51
• Top Feature: Total Area (71.5%), Orientation more important than in monsoon

Insight: Summer electricity consumption is more influenced by occupant behavior and solar exposure, while monsoon usage is dominated by structural factors.

---

Model Diagnostics

Residual analysis was performed for both monsoon and summer models. Monsoon residuals show stable, random error distribution, while summer residuals exhibit higher variance due to unobserved behavioral factors.

How to Run

1. Install requirements:

pip install -r requirements.txt