Pime Diabetes Prediction

Overview

This project implements a complete machine learning pipeline to predict diabetes risk using the Pima Indians dataset. It covers data preprocessing, model building, validation, optimization, and deployment via a FastAPI-based API.

Content:

• Scripts
• Notebooks
• Deployment

Online Project
For more details and updates about this project, visit this website.

---

Scripts

data_preprocessing.py

• Loads and transforms raw data (replacing zero values with NaN for key features).
• Generates 8 preprocessing combinations (applying/removing outliers, balancing, scaling).
• Saves processed CSV files with descriptive names.

pipeline.py

• Configures preprocessing using simple and iterative imputations via a ColumnTransformer.
• Integrates a pre-trained classifier into a unified pipeline.
• Trains the model using processed data and saves the pipeline.

---

Notebooks

• 01_data_preprocessing.ipynb: Data import, cleaning, exploration, and normalization.
• 02_data_model_validation.ipynb: Splitting data, model evaluation with metrics, and comparison.
• 03_model_optimization.ipynb: Hyperparameter tuning and robust model validation.

Diagrams illustrate the ML pipeline architecture and best model performance.

---

Deployment

• API Overview:
  A FastAPI-based service for diabetes prediction.

• Key Endpoints:

  • GET /: Health check.
  • POST /predict: Returns a binary prediction.
  • POST /predict_proba: Returns class probability scores.

• Deployment Methods:
  Includes instructions for local Docker deployment and scalable deployment on AWS Elastic Beanstalk.

---

Contributing

We welcome contributions! If you have ideas, improvements, or fixes, please feel free to fork the repository and submit pull requests. Your involvement helps make this project even better.