Electricity Theft Detection System

Purpose

Electricity theft is a major issue in regions like Karachi, where unauthorized consumption of electricity leads to significant losses for utility companies. This project provides a solution for detecting electricity theft using machine learning models. By analyzing various factors such as electricity usage, voltage fluctuations, and historical data, the system predicts the likelihood of theft.

Model Overview

The machine learning model used in this project is a Random Forest Classifier. This model was chosen due to its ability to handle complex data with multiple features and its robustness in classification tasks.

Key Techniques Used:

1. SMOTE (Synthetic Minority Over-sampling Technique): Used to handle class imbalance in the dataset. It generates synthetic samples for the underrepresented class (theft) to improve model performance.

2. Grid Search for Hyperparameter Tuning: The model was fine-tuned using GridSearchCV, which optimizes the hyperparameters of the Random Forest classifier to find the best-performing configuration. Key hyperparameters tuned include:

   • n_estimators: The number of trees in the forest (100, 200, 300).
   • max_depth: The maximum depth of the trees (None, 10, 20).
   • min_samples_split: The minimum number of samples required to split an internal node (2, 5, 10).

Model Pipeline:

The model is part of a pipeline that includes:

• Standard Scaling: Scales input features to standardize the dataset.
• Random Forest Classifier: A robust classification model to predict the likelihood of electricity theft.

How to Run the Streamlit App

1. Install Dependencies:
   Before running the Streamlit app, make sure to install all necessary dependencies. You can do this by running the following command:

   pip install -r requirements.txt
   

2. Run the Streamlit App:
   Once the dependencies are installed, you can run the Streamlit app using the following command:

streamlit run app.py

This will launch the app in your web browser, typically at http://localhost:8501.