Fraud Detection Project

This project analyzes and detects fraudulent transactions using a dataset of financial transactions. The workflow includes data exploration, visualization, feature engineering, and building a machine learning model to classify transactions as fraudulent or not. The trained model is deployed using Streamlit for interactive visualization and prediction.

Dataset

• The dataset is loaded from an Excel file (Fraud.xlsx).
• Key columns include transaction type, amount, balances, and fraud indicators.

Steps Performed

1. Data Exploration & Cleaning

   • Checked for missing values and data types.
   • Explored class distribution for isFraud and isFlaggedFraud.
   • Visualized transaction types and fraud rates.

2. Feature Engineering

   • Created new features such as balance differences.
   • Filtered and analyzed suspicious patterns (e.g., zero balances after transfer).

3. Visualization

   • Plotted distributions of transaction amounts.
   • Visualized fraud rates by transaction type and over time.
   • Correlation heatmaps for key features.

4. Model Building

   • Selected features and split data into training and test sets.
   • Preprocessed data using scaling and one-hot encoding.
   • Built a pipeline with logistic regression (class weight balanced).
   • Evaluated model with classification report and confusion matrix.

5. Model Saving

   • Saved the trained pipeline using joblib for future use.

6. Deployment with Streamlit

   • Developed a Streamlit app for interactive visualization and prediction.
   • Users can upload transaction data and get real-time fraud predictions.
   • Visualizations of transaction patterns and model results are available in the app.

Model Performance

• Accuracy: ~ 94%

Requirements

• Python 3.x
• pandas
• numpy
• matplotlib
• seaborn
• scikit-learn
• joblib
• streamlit

Install dependencies with:

pip install pandas numpy matplotlib seaborn scikit-learn joblib streamlit

Usage

1. Place Fraud.xlsx in the specified directory.
2. Run the notebook Fraud_Detection.ipynb step by step to train and save the model.
3. Start the Streamlit app:

   streamlit run app.py

4. Use the web interface to visualize data and make predictions.

Results

• The model provides classification metrics and a confusion matrix for fraud detection.
• Visualizations help understand transaction patterns and fraud distribution.
• The Streamlit app allows for interactive exploration and prediction.

License

This project is for educational purposes.