InsureSight AI

InsureSight AI is an ML powered app for predicting health insurance premiums based on personal & medical profiles.

Frontend: Streamlit   |   ML Framework: XGBoost & scikit-learn   |   Language: Python

This project uses XGBoost regression models to predict health insurance premiums based on age, medical history, lifestyle factors, & demographics.

---

A machine learning web application that predicts health insurance premium costs based on user-provided personal and medical information. The application features an intuitive Streamlit interface, age-specific model selection, risk score normalization, and real-time predictions.

---

Features

• Age-Based Model Selection: Uses separate trained models for users aged ≤25 and >25 for improved accuracy.
• Comprehensive Risk Assessment: Calculates normalized risk scores based on medical history conditions (Diabetes, Heart Disease, Blood Pressure, Thyroid).
• Feature Engineering: Automatically encodes categorical variables (Gender, Region, BMI Category, Smoking Status, etc.).
• Real-Time Predictions: Instant premium estimates displayed in INR format.
• Clean UI: Organized 4-row grid layout with intuitive icons for all input fields.
• Input Validation: Enforces valid ranges for all numeric inputs (Age, Dependants, Income, Genetic Risk).
• Pre-Trained Models: Pre-trained XGBoost models and scalers loaded via joblib for fast inference.

---

📸 Screenshots

🖥️ Streamlit Application Interface

✅ Sample Prediction Results

---

Folder Structure

insurance_premium_prediction/
│
├── artifacts/
│   ├── model_u25.joblib          # XGBoost model for age ≤ 25
│   ├── model_others.joblib       # XGBoost model for age > 25
│   ├── scaler_u25.joblib         # Feature scaler for age ≤ 25
│   └── scaler_others.joblib      # Feature scaler for age > 25
│
├── main.py                       # Streamlit frontend application
├── prediction_helper.py          # Prediction logic & preprocessing
├── requirements.txt              # Project dependencies
├── .gitignore                    # Git ignore file
├── LICENSE                       # License file
└── README.md                     # Project documentation

---

Setup Instructions

Prerequisites

• Python 3.10 or higher
• Git

1. Clone the Repository

git clone https://github.com/inv-fourier-transform/insurance-premium-prediction.git
cd insurance-premium-prediction

2. Create and Activate a Virtual Environment

python -m venv venv
# On Windows:
venv\Scripts\activate
# On macOS/Linux:
source venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

Running the Application

streamlit run main.py

Input Features

The application collects the following information to generate premium predictions:

Demographics & Personal Info

• Age (18–100): Primary insured's age
• Gender: Male or Female
• Marital Status: Married or Unmarried
• Number of Dependants (0–10): Family members covered

Financial & Employment

• Income in Lakhs (0–200): Annual income in Indian Rupees (Lakhs)
• Employment Status: Salaried, Self-Employed, or Freelancer

Health & Lifestyle

• BMI Category: Normal, Overweight, Obesity, or Underweight
• Smoking Status: No Smoking, Occasional, or Regular
• Genetical Risk (0–5): Genetic predisposition risk score
• Medical History:
  • No Disease
  • Diabetes
  • High Blood Pressure
  • Heart Disease
  • Thyroid
  • Or combinations of the above

Coverage Details

• Insurance Plan: Bronze, Silver, or Gold coverage tier
• Region: Northwest, Southeast, Northeast, or Southwest

---

How It Works

1. Risk Score Calculation

The calculate_normalized_risk() function in prediction_helper.py assigns risk scores to medical conditions:

• Heart Disease: 8
• Diabetes / High Blood Pressure: 6
• Thyroid: 5
• No Disease: 0

Combined conditions (e.g., Diabetes & Heart Disease) are summed and normalized to a 0–1 scale.

---

2. Feature Preprocessing

• Categorical Encoding:
  One-hot encoding for Gender, Region, Marital Status, BMI Category, Smoking Status, and Employment Status

• Ordinal Encoding:
  Insurance Plan → Bronze = 1, Silver = 2, Gold = 3

• Income Level Mapping:
  Income converted to categorical levels:
  <10L, 10L–25L, 25L–40L, >40L

---

3. Age-Based Scaling & Prediction

• Users aged ≤ 25 use scaler_u25 and model_u25
• Users aged > 25 use scaler_others and model_others

Features are scaled using the appropriate scaler before prediction.
The XGBoost regression model returns the premium estimate.

---

Model Training (Optional)

To retrain the models with your own data:

1. Prepare your dataset with the features listed above and target variable premium_amount
2. Split data by age (≤25 and >25)
3. Train separate XGBoost regressors for each age group
4. Save models and scalers to the artifacts/ directory using joblib.dump()

---

Contributing

Contributions are welcome!
Please open an issue or submit a pull request for improvements, bug fixes, or feature enhancements.

---

License

This project is licensed under the MIT License.

---

🎯 Predict your health insurance premiums with confidence using machine learning!