🔍 Insurance Fraud Detection

This project aims to detect fraudulent insurance claims using machine learning techniques. With a dataset of real-world insurance claims, the model identifies patterns and anomalies that help in flagging potentially fraudulent activities.

📌 Problem Statement

Insurance fraud leads to significant financial losses each year. Identifying fraud early can save costs and ensure better claim verification processes. This project builds a model to distinguish between genuine and fraudulent claims using classification algorithms.

📊 Dataset

The dataset used contains various features related to insurance claims including:

• Demographics (e.g., Age, Gender)
• Claim history (e.g., Number of past claims)
• Policy details (e.g., Policy type, Deductible)
• Incident details (e.g., Accident Area, Fault)
• Target: FraudFound (1: Fraudulent, 0: Not Fraudulent)

🛠️ Technologies Used

• Python
• Pandas, NumPy – Data manipulation
• Seaborn, Matplotlib – Data visualization
• LabelEncoder – Categorical encoding
• XGBoost – Feature importance and modeling
• SMOTE – Class imbalance handling
• Scikit-learn – Model training and evaluation

🧠 Workflow

1. Data Cleaning: Removed irrelevant columns and handled missing values.
2. Exploratory Data Analysis (EDA): Visualized fraud trends across various categories.
3. Encoding: Converted categorical variables using Label Encoding.
4. Feature Selection: Used XGBoost to identify top features contributing to fraud detection.
5. SMOTE: Applied Synthetic Minority Oversampling Technique to balance classes.
6. Model Training: Trained an XGBoost Classifier on the balanced dataset.
7. Evaluation: Compared model performance before and after SMOTE using accuracy and class distribution.

📈 Results

• Successfully balanced the dataset using SMOTE.
• Top contributing features included: Fault, PolicyType, AddressChange-Claim, AccidentArea, etc.
• XGBoost model provided interpretability and effective feature ranking.

✅ Conclusion

This project shows how machine learning can be leveraged to flag potentially fraudulent claims and assist insurance companies in early detection, reducing financial losses.

📂 Folder Structure

insurance-fraud-detection/ │ ├── insurance_fraud_detection.ipynb # Jupyter Notebook with full project ├── carclaims.csv # Dataset (if shared) ├── README.md # Project documentation

🙌 Acknowledgments

• Dataset: Kaggle
• Libraries: Scikit-learn, XGBoost, imbalanced-learn

---