| --- |
| license: mit |
| tags: |
| - machine-learning |
| - fraud-detection |
| - classification |
| - decision-tree |
| - scikit-learn |
| library_name: scikit-learn |
| --- |
| |
| # 🛡️ Insurance Fraud Detection |
|
|
| A machine learning classification model for predicting whether an insurance claim is **Fraudulent** or **Non-Fraudulent**. |
|
|
| The model uses a **Decision Tree Classifier** trained on insurance claim data after preprocessing, feature engineering, categorical encoding, and class balancing using SMOTE. |
|
|
| ## 🤖 Model |
|
|
| **Algorithm:** Decision Tree Classifier |
|
|
| **Task:** Binary Classification |
|
|
| **Classes:** |
|
|
| - Fraudulent |
| - Non-Fraudulent |
|
|
| ## 📊 Model Performance |
|
|
| The following models were evaluated during the project: |
|
|
| | Model | Accuracy | |
| |---|---:| |
| | Logistic Regression | 73% | |
| | Decision Tree | **80%** | |
| | Random Forest | 79% | |
|
|
| The Decision Tree Classifier was selected for the final application and achieved approximately **80% accuracy** on the evaluation data. |
|
|
| > Accuracy alone does not fully describe fraud-detection performance. Precision, recall, F1-score, and the confusion matrix should also be considered, particularly when the classes are imbalanced. |
|
|
| ## 🔄 Machine Learning Workflow |
|
|
| ```text |
| Insurance Claim Data |
| ↓ |
| Data Cleaning |
| ↓ |
| Missing Value Handling |
| ↓ |
| Feature Engineering |
| ↓ |
| Categorical Encoding |
| ↓ |
| SMOTE Class Balancing |
| ↓ |
| Decision Tree Classifier |
| ↓ |
| Fraud / Non-Fraud Prediction |
