⚡ Energy Theft Detection with Yearly Average Feature & Dashboard

Detect potential electricity theft using Random Forest and yearly average consumption trends. Includes a dashboard to classify customer risk levels and highlight high-risk cases.

🔹 Key Features

• Generate or load balanced synthetic dataset for Normal, Medium, and High-risk customers.

• Compute yearly average consumption per customer for improved model accuracy.

• Random Forest classifier to detect potential theft.

• Vacation adjustment to reduce false positives.

• Dashboard visualization including:

  • Customer risk distribution
  • Top high-risk customers
  • Feature importance

📊 Dataset

• File: energy_balanced_data.csv
• Columns:

Column	Description
customer_id	Unique customer ID
month	Month (1–12)
consumption_kwh	Monthly electricity consumption
is_theft	Label: 0 = Normal, 1 = Potential Theft
yearly_avg	Computed yearly average consumption

Dataset is synthetic but modeled to mimic real-world electricity usage.

Download Dataset

⚡ Quick Start

1. Clone the repository:

git clone https://github.com/<your-username>/energy-theft-detection.git
cd energy-theft-detection

2. Install dependencies:

pip install -r requirements.txt

3. Run the main script:

python energy_theft_detection.py

---

🛠 How It Works

1. Load or generate dataset – 1000+ customers, 12 months each.
2. Feature Engineering – Compute yearly_avg per customer.
3. Train Model – Random Forest (300 trees).
4. Evaluate Model – Accuracy, confusion matrix, classification report, feature importance.
5. Predict New Customer – Input monthly consumption, yearly average, vacation status.
6. Dashboard – Visualize risk levels and top high-risk customers.

📈 Dashboard & Visualization

• Risk Distribution: Normal, Medium, High Risk
• Top High-Risk Customers: Sorted by theft probability
• Feature Importance: Contribution of month, consumption, yearly average

Sample Dashboard:

🧪 Example Prediction

text Month: 5 Consumption: 180 kWh Yearly Average: 300 kWh Vacation: yes

Output: ⚠️ Alert! Possible energy theft (Probability: 0.75)

• Threshold for High Risk: 0.7
• Vacation adjustment reduces probability to avoid false positives

💾 Save & Load Model

python

Save model

joblib.dump(model, "energy_theft_yearlyavg_model.pkl")

Load model

model = joblib.load("energy_theft_yearlyavg_model.pkl")

📊 Results

• Accuracy: Replace with your model score
• Classification Report: Precision, Recall, F1-score for Normal vs Theft
• Feature Importance Ranking: consumption_kwh > yearly_avg > month

🚀 Future Enhancements

• Integrate real electricity consumption datasets.
• Add seasonal, regional, and holiday adjustments.
• Build an interactive web dashboard using Streamlit or Dash.
• Include unsupervised anomaly detection for rare theft patterns.

📷 Screenshots

Screenshot	Description
	Model performance evaluation
	Feature impact on prediction
	Customer risk classification

📄 License

MIT License – See LICENSE

Author-TejaHussain💖