🩺 Obesity Prediction Project

This project aims to predict obesity levels based on individuals' health metrics and lifestyle behaviors. The workflow includes preprocessing, feature engineering, model building, evaluation, and deployment via a Streamlit web interface.

---

🌐 Live Demo

🖥️ Try the application here:
👉 https://obesity-test.streamlit.app/

🖥️Read the full article on Medium:

👉 https://medium.com/@sercanteyhani/predicting-obesity-with-machine-learning-from-raw-data-to-real-time-web-app-5e601f4267e5

---

📌 Project Workflow

Obesity Prediction Using Machine Learning

---

1. Project Overview

This project predicts obesity levels using machine learning, deploying the model as a real-time web application.

---

2. Dataset and Preprocessing

• Verified the dataset is balanced across obesity classes.
• Applied Label Encoding to transform categorical variables.
• Split the data into 80% training and 20% testing sets.
• Scaled features using StandardScaler for normalization.

---

3. Model Training

• Used Logistic Regression with the newton-cg solver.
• Applied class_weight='balanced' to handle class imbalance.
• Trained the model on the training dataset.
• Evaluated the model with:
  • Classification report (precision, recall, F1-score)
  • Log Loss metric
  • Confusion matrix for detailed error analysis

---

4. Feature Engineering - BMI Creation

• Created a new feature: BMI (Body Mass Index) using the formula:
  BMI = Weight (kg) ÷ (Height (m))²
• Incorporated BMI into the dataset to better capture obesity patterns.
• Dropped the original Weight and Height columns to avoid redundancy and multicollinearity.

---

5. Feature Importance and Explainability

• Applied SHAP (SHapley Additive exPlanations) to interpret feature impact.
• Visualized feature importance using SHAP summary plots.
• Some features were removed based on their low importance to improve model simplicity and performance.

---

6. Model Retraining

• The model was retrained to improve performance and ensure balanced class handling.

---

7. Model Saving

• Saved the trained model using joblib for later use in the Streamlit app.

---

8. Streamlit Interface

• Developed a user-friendly Streamlit web app where users can input relevant personal data and receive obesity level predictions in real time.

---

🧰 Technologies Used

• Python: Data manipulation and modeling
• Pandas / NumPy: Data preprocessing
• Matplotlib / Seaborn: Visualizations
• Scikit-learn: Model building, scaling, evaluation, Confusion matrix & classification metrics, Feature importance analysis
• Joblib: Model persistence
• Streamlit: Web deployment interface

📸 Project Screenshots