🏗️ BuildCast - Building Energy Prediction

This project is a machine learning pipeline for predicting building energy usage using synthetic datasets.
It includes data preprocessing, model training (LightGBM + PyTorch), evaluation, and visualization of results.

---

📌 Features

• Preprocessing with StandardScaler and OneHotEncoder
• LightGBM baseline model for regression
• PyTorch neural network model implementation
• Evaluation metrics: MAE, RMSE, R²
• Graphical visualization of predicted vs actual values
• Synthetic dataset generation with 20,000 data points for better visualization

---

📂 Project Structure

📦 Building-Energy-Prediction

• 📜 main.py # Main training + evaluation pipeline
• 📜 requirements.txt # Dependencies
• 📜 buildings.csv # Example dataset
• 📜 README.md # Project documentation

---

⚙️ Installation

1. Clone the repository:

git clone https://github.com/QuantumCoderrr/BuildCast.git
cd BuildCast

2. Create a virtual environment & install dependencies:

python -m venv venv
source venv/bin/activate   # On Mac/Linux
venv\Scripts\activate      # On Windows

pip install -r requirements.txt

---

▶️ Usage

1. Run with existing dataset

python main.py

2. Generate & use synthetic dataset (20,000 points)

python generate_dataset.py
python main.py

---

📊 Visualization

The model outputs a scatter plot comparing actual vs predicted energy usage. This helps visualize how well the model generalizes on test data.

📈 Example Output

Metrics (MAE, RMSE, R²) printed on console Scatter plot of predictions vs actual values

🛠️ Tech Stack

Python 3.11+ Pandas, NumPy Scikit-learn LightGBM PyTorch Matplotlib

---

👩‍💻 Author

Sandip Ghosh, Aishika Majumdar and Sandhita Poddar

---

📜 License

This project is licensed under the MIT License. Feel free to use, modify, and share with proper attribution.