🛡️ Insurance Claim Fraud Detection

This project detects fraudulent insurance claims using machine learning models. It includes data preprocessing, exploratory data analysis (EDA), feature engineering, model training, and deployment with a simple Flask web app.

📌 Features

    Cleaned and preprocessed insurance claim dataset.
    Exploratory Data Analysis (EDA) with visualizations.
    Machine learning models trained for fraud detection.
    Streamlit web app for real-time fraud prediction.

    Input transaction can be saved as raw + transformed CSV for auditing.

📂 Project Structure

        ├── app.py                                  # Streamlit application
        ├── training_encoders.py                    # training categorical columns 
        ├── biased fraud-like insurance claims.py   # Fraud-like case generated 
        ├── model/                                  # Saved ML models 
        ├── PreProcessing/                          # Saved encoders
        ├── notebooks/                              # EDA & model building notebooks
        ├── data/                                   # Dataset folder 
        ├── Images/                                 # snipped from Streamlit application 
        ├── synthetic_fraud_case_csv/               # saved Fraud-like case generated  
        ├── requirements.txt                        # Project dependencies
        └── README.md                               # Project documentation

⚙️ Installation

Clone the repo

    git clone https://github.com/Decadent-tech/EDA-on-Insurance-Claim-Fraud-Detection
    cd EDA-on-Insurance-Claim-Fraud-Detection

Create a virtual environment and install dependencies

    pip install -r requirements.txt

Run the Streamlit app

    streamlit run app.py                                                          


    Open in browser: http://localhost:8501/

🖥️ Usage

    Enter 'months_as_customer','policy_state','policy_csl','policy_deductable',
    'policy_annual_premium','umbrella_limit','insured_zip','insured_sex',
    'insured_education_level','insured_occupation','insured_hobbies',
    'insured_relationship','capital-gains','capital-loss','incident_type',
    'collision_type','incident_severity','authorities_contacted',
    'incident_state','incident_city','incident_hour_of_the_day',
    'number_of_vehicles_involved','property_damage','bodily_injuries',
    'witnesses','police_report_available','total_claim_amount',
    'injury_claim','property_claim','auto_make','auto_year',
    'incident_month','incident_day','auto_model'.

    The model will output whether the claim is fraudulent (1) or genuine (0).

    Both raw and transformed inputs are saved in CSV for auditing. # Auditing pupose 

📊 Models Used

    XGBoost
    Evaluation metrics: Accuracy, Precision, Recall, F1-Score, ROC-AUC

📸 Screenshots

🔹 Web App UI

🚀 Future Enhancements

Add model explainability (SHAP/LIME).
Deploy on Render / Hugging Face Spaces.
Containerize using Docker + Docker Compose.
CI/CD pipeline with GitHub Actions.

🙌 Author

Debo