🔋 Household Energy Consumption Forecasting

This project aims to predict Global Active Power (kilowatts) consumed in a household using various machine learning models. The dataset used is the Household Power Consumption Dataset.

---

📂 Dataset Overview

• Source: UCI Machine Learning Repository
• Rows: ~2 million
• Target Variable: Global_active_power
• Features:
  • Global_reactive_power
  • Voltage
  • Global_intensity
  • Sub_metering_1, Sub_metering_2, Sub_metering_3

---

🧹 Data Preprocessing

• Missing values handling
• Feature engineering (timestamp extraction, interaction features)
• Normalization / Scaling (StandardScaler)
• Train-test split (typically 80/20)

---

🔧 Models Implemented

1. 📈 Linear Regression

• Simple baseline model
• Assumes linear relationship between input features and target
• Easy to interpret, but prone to underfitting

2. 🌳 Decision Tree Regressor

• Captures nonlinear relationships
• May overfit on training data
• Visualized using plot_tree()

3. 🌲 Random Forest Regressor

• Ensemble of multiple decision trees
• Reduces overfitting and improves generalization
• Hyperparameter tuning: n_estimators, n_jobs, etc.

4. 🚀 Gradient Boosting Regressor

• Boosts weak learners sequentially
• Slower than Random Forest but potentially more accurate
• Hyperparameters tuned via GridSearchCV

5. ⚡ XGBoost Regressor

• Extreme Gradient Boosting (high-performance)
• Handles missing values internally
• Feature importance visualization using xgboost.plot_importance()

6. 🧠 Artificial Neural Network (ANN)

• Deep learning model using TensorFlow/Keras
• Architecture:
  • Input layer with 14 features
  • Hidden layers with ReLU activations
  • Output layer with 1 neuron (linear activation)
• Optimizer: Adam | Loss: MSE | Metric: MAE

---

📊 Evaluation Metrics

• Mean Absolute Error (MAE)
• Mean Squared Error (MSE)
• R² Score

Model	MAE	MSE	R² Score
Linear Regression	0.104	0.065	0.938
Decision Tree	0.031	0.022	0.980
Random Forest	0.025	0.001	0.998
Gradient Boosting	0.051	0.031	0.972
XGBoost	0.041	0.021	0.981
ANN	0.138	0.097	0.912

---

🖼️ Visualizations

• Correlation Heatmaps
• Feature Importance Plots
• Model Prediction vs Actual Line Charts
• Residual Plots

---

📦 Dependencies

pandas
numpy
matplotlib
seaborn
scikit-learn
xgboost
tensorflow / keras