Energy Consumption Forecasting 🏠⚡

This repository contains a comprehensive time series forecasting project on household electricity consumption (Global Active Power) using SARIMA, Prophet, and LSTM (PyTorch).

---

📁 Project Structure

• notebook.ipynb – Jupyter notebook with full analysis: data cleaning, EDA, modeling, forecasting, and evaluation.
• README.md – This file.
• (Data not included for size/license reasons.)

---

📊 Data Source

The dataset originates from the UCI Machine Learning Repository and is hosted on Kaggle:
"Household Electric Power Consumption" — electric power consumption measurements in one household at one-minute intervals over nearly four years (December 2006–November 2010) :contentReference[oaicite:1]{index=1}.

---

🧰 Tools & Libraries Used

• pandas, numpy, matplotlib, seaborn – for data manipulation and visualization
• statsmodels – to implement SARIMA models
• fbprophet (Prophet) – for automated trend and seasonality forecasting
• scikit-learn – for scaling and error metrics (RMSE, MAE)
• torch (PyTorch) – for building and training the LSTM neural network

---

🛠️ What’s in the Notebook?

1. Data Loading & Cleaning – parsing date/time, handling missing values, resampling.
2. Exploratory Data Analysis (EDA) – global daily plots, autocorrelation, seasonality insights.
3. Forecasting Methods:
   • SARIMA (seasonal ARIMA) with parameter grid-search using AIC,
   • Prophet (robust to trend shifts and missing data),
   • LSTM (deep learning-based sequence modeling using PyTorch).
4. Model Evaluation – Forecast vs. actual plots; performance compared using RMSE and MAE at hourly and daily aggregation.
5. Conclusion & Lessons Learned – interpretation, strengths/weaknesses, and suggestions for future extension (e.g. external features).

---

✅ Usage Instructions

To replicate this project:

1. Download the dataset from Kaggle: Household Power Consumption Dataset.
2. Unzip and place household_power_consumption.txt in the repository root directory.
3. Open and run Time_Series.ipynb in Jupyter.

📌 Key Learnings

• Forecasting on real-world time series requires careful handling of missing data and seasonality.
• Prophet consistently outperformed SARIMA and LSTM in both accuracy and forecasting realism.
• LSTM demonstrated potential but required more data preprocessing and tuning.
• Visual and metric-based model comparison (e.g., RMSE, MAE, daily vs. hourly) is essential to assess real-world usability.

---

This project demonstrates hands-on capability in time series forecasting methods, thoughtful data analysis, and clear evaluation—ideal for applications in energy analytics, demand planning, or predictive modeling roles.

---

License: MIT License

Enjoy exploring and feel free to reach out with any questions or feedback!