📘 Building-Level Electricity Access and Demand Analysis

Authors: Pavan Kumar Yadav
Institution: Government Engineering College, Dahod

---

📌 Introduction

Buildings represent a major portion of global energy consumption. This project focuses on analyzing building-level electricity access and demand to predict monthly electricity usage and classify buildings based on consumption levels. Such analysis plays a crucial role in improving operational efficiency, energy planning, and sustainability.

The integration of AI-based systems for analyzing real-time occupancy and environmental data can significantly reduce energy waste and operational costs. This approach mirrors successful implementations in smart city and energy optimization initiatives worldwide.

---

📊 Dataset Description

This study utilizes high-resolution building-scale energy access and demand datasets provided by the International Energy Agency (IEA).

🔑 Key Features

• Building Area: Footprint size of each building (in square meters)
• Electricity Access: Percentage of electricity accessibility per building
• Nearby Building Density: Number of buildings within a 1 km radius
• Target (Regression):
  • cons (kWh/month) – Monthly electricity consumption
• Derived Target (Classification):
  • Energy Consumption Level: Low / Medium / High

📂 Data Source

• IEA Building-Level Electricity Access and Demand Model for Sub-Saharan Africa
• Files used:
  • 471928_geoms.csv
  • 466151_geoms.csv

Note:
Although the datasets are in CSV format, they contain a geometry column with embedded GeoJSON spatial data, which is used to identify building footprints and spatial relationships.

---

🛠️ Methodology & Performance

The project follows a complete machine learning pipeline:

1. Data Cleaning & Preprocessing
2. Feature Engineering
3. Regression Modeling
4. Ordinal Classification

🔍 Models Used

• Linear Regression
• Random Forest
• XGBoost

📈 Key Findings

• Best Model: XGBoost
• Classification Accuracy: ~98.9%
• Model Robustness:
  • Both Random Forest and XGBoost achieved high precision, recall, and F1-scores across all energy consumption categories.
• Ordinal Analysis:
  • Energy consumption categories are ordinal in nature (Low < Medium < High).
  • Evaluation metrics and modeling strategies were aligned with this ordered structure to improve statistical stability and interpretability.

---

📜 License & Attribution

This dataset is released under the Open Database License (ODbL), Version 1.0.

Attribution:
© International Energy Agency (IEA).
Data sourced from the Building-level electricity access and demand model.

---

💻 Setup & Execution

This project is designed for execution in a Google Colab environment.

⚙️ Steps to Run

1. Upload the IEA CSV files to the Colab runtime:

   • 471928_geoms.csv
   • 466151_geoms.csv

2. Install / import required libraries:

   import pandas as pd
   import numpy as np
   import matplotlib.pyplot as plt
   import seaborn as sns
   from sklearn.model_selection import train_test_split
   
   from sklearn.metrics import classification_report
   from sklearn.ensemble import RandomForestClassifier

3. Run the preprocessing, regression, and classification notebooks/scripts.

✅ Conclusion

This project demonstrates how building-level data combined with machine learning can effectively predict electricity demand and classify consumption patterns. The results highlight the potential of AI-driven energy analytics for smarter urban planning, energy efficiency, and sustainable infrastructure development.