Life Expectancy Predictor

Live Demo: https://life-expectancy-jvcx.onrender.com/

A machine learning project that predicts life expectancy (in years) using health and socioeconomic indicators.
A Random Forest model achieves R² = 0.958 (RMSE = 1.93) and is served via a FastAPI web application.

---

📌 Overview

This project answers the question:
“Given a set of country‑level health and economic factors, what is the expected life expectancy?”

I worked on a cleaned and feature‑engineered version of the WHO Global Health Observatory data. After exploratory analysis and feature selection, I trained several regression models and selected a Random Forest for its high accuracy.
The model is deployed as a REST API with a simple HTML frontend, allowing users to input values and obtain an instant prediction.

---

📂 Dataset

• Source: WHO Global Health Observatory / Kaggle
• Original file: life_expectancy_data.csv – 2938 rows × 22 columns
• Cleaned file: life_expectancy_data_cleaned.csv
• Processed file (used for modelling): life_expectancy_data_processed.csv

Columns include:

• Target: life_expectancy
• Health indicators: adult_mortality, infant_deaths, under_five_deaths, hiv/aids, bmi, hepatitis_b, measles, polio, diphtheria, alcohol, percentage_expenditure, total_expenditure
• Socioeconomic: schooling, income_composition_of_resources, gdp, population, status (Developing/Developed), thinness_1_19_years, thinness_5_9_years

Data Cleaning & Preparation

1. Handled missing values – imputed with median for numeric columns, and mode for categorical.
2. Corrected data types and fixed inconsistent entries.
3. Outlier detection using IQR method, but retained because the data points reflected genuine extremes.
4. Created a cleaned version (life_expectancy_data_cleaned.csv).

---

🔍 Exploratory Data Analysis (EDA)

I performed EDA using both Python (pandas, matplotlib, seaborn) and SQL to understand relationships:

• Life expectancy vs. Schooling – strong positive correlation (r ≈ 0.75)
• Life expectancy vs. Income composition of resources – strong positive correlation
• Life expectancy vs. HIV/AIDS – negative correlation, especially in Sub‑Saharan Africa
• Adult mortality and BMI also showed significant non‑linear patterns

Key insights from EDA guided feature engineering and model selection.

---

🧪 Feature Engineering & Selection

To capture non‑linearities and interactions, I created derived features:

1. income_comp_sq – squared income composition of resources
2. log_hiv_aids – log(1 + HIV/AIDS) to reduce skewness
3. schooling_sq – squared schooling
4. bmi_sq – squared BMI
5. mortality_per_bmi – adult_mortality / (bmi + 1) (interaction)
6. mortality_hiv_inter – adult_mortality * hiv/aids (interaction)

I retained:

• adult_mortality
• status (categorical: 0 = Developing, 1 = Developed)

Features after selection: a total of 8 predictors.

VIF (Variance Inflation Factor) was checked to ensure no multicollinearity; all selected features had VIF < 10.

Finally, I scaled the continuous features using StandardScaler (saved as scaler.joblib).

---

🤖 Modelling

I trained and evaluated multiple regression models:

Model	RMSE (years)	MAE (years)	R² Score
Dummy Regressor	9.45	7.76	0.0
Linear Regression	3.71	2.74	0.848
Lasso Regression	3.63	2.69	0.852
Ridge Regression	3.62	2.69	0.852
Random Forest	1.93	1.20	0.958
XGBoost	2.00	1.27	0.955

Random Forest was chosen as the final model because of its high accuracy, robustness, and ability to capture non‑linear relationships.

The trained model is saved as random_forest_model.pkl and the scaler as scaler.joblib.

---

🌐 Web Application

A FastAPI web app serves the model:

• Backend: api.py – REST endpoint /predict that accepts JSON input, engineers features, scales them, and returns the predicted life expectancy.
• Frontend: index.html – a simple UI where users can enter values and see the prediction instantly.

Example API request:

{
  "income_composition_of_resources": 0.6,
  "hiv_aids": 0.1,
  "schooling": 12.0,
  "adult_mortality": 150.0,
  "bmi": 25.0,
  "status": 0
}

Response:

{
  "predicted_life_expectancy": 72.45,
  "units": "years"
}

🚀 How to Run Locally

1. Clone the repository

   git clone git@github.com:MohammadErfanRashidi/Life-Expectancy-.git
   cd life-expectancy-predictor

2. Create a virtual environment & install dependencies

   python -m venv venv
   source venv/bin/activate   # On Windows: venv\Scripts\activate
   pip install -r requirements.txt

3. Run the API server

   uvicorn api:app --reload

4. Open the app

   • Frontend: http://127.0.0.1:8000
   • API docs (Swagger): http://127.0.0.1:8000/docs

📁 Project Structure

.
├── api.py
├── data
│   ├── life_expectancy_data_cleaned.csv
│   ├── life_expectancy_data.csv
│   └── life_expectancy_data_processed.csv
├── index.html
├── LICENSE
├── model
│   ├── feature_names.json
│   ├── random_forest_model.pkl
│   └── scaler.joblib
├── notebooks
│   ├── cleaning.ipynb
│   ├── EDA.ipynb
│   ├── modeling.ipynb
│   └── preprocessing.ipynb
├── README.md
├── reports
│   ├── python_EDA_viz
│   │   ├── correlation.png
│   │   ├── distribution.png
│   │   ├── life_expectancy_overtime.png
│   │   ├── pairplot.png
│   │   ├── scatterplot.png
│   │   └── status.png
│   └── sql_eda_results
│       ├── adult_mortality_on_life.csv
│       ├── alcohol_on_life.csv
│       ├── gdp_on_life.csv
│       ├── icr_on_life.csv
│       ├── infant_deaths_on_life.csv
│       ├── schooling_on_life.csv
│       └── status_per_year_on_life.csv
├── requirements.txt
├── sql_EDA
│   └── EDA.sql
└── src

🛠️ Technologies Used

• Python 3.9+
• pandas, NumPy, scikit‑learn, XGBoost
• Matplotlib, Seaborn (EDA)
• SQL (data exploration)
• FastAPI, Uvicorn, Pydantic (web API)
• HTML/CSS/JavaScript (frontend)

📈 Results & Conclusion

The Random Forest model explains 95.8% of the variance in life expectancy, meaning the selected features (income, schooling, HIV rate, adult mortality, BMI and their interactions) are extremely predictive.
This project demonstrates end‑to‑end machine learning: from data cleaning, EDA, feature engineering, model tuning, to API deployment.

📄 License

MIT License. See LICENSE for details.