Project Overview

End-to-end pipeline for diabetes prediction using Neural Networks in both Keras and PyTorch. This project covers:

• Exploratory Data Analysis & Feature Engineering
• Advanced Preprocessing: SMOTE for class balance, PCA for dimensionality reduction, polynomial feature expansion
• Model Definitions: Separate implementations in Keras and PyTorch
• Hyperparameter Tuning: Automated search with Keras Tuner
• Training & Evaluation: Code to train, validate, and compare models across frameworks

Features

• Streamlined data ingestion and preprocessing
• Modular model architectures you can extend
• Automated Keras hyperparameter sweeps
• Clear training/validation routines with accuracy and loss reporting

Data

• pima-indians-diabetes.csv Contains the original database without any preprocessing.
• Diabetes_1.csv Contains the base that is reached after feature engineering and data imputation

Important Dependencies

pip install pandas numpy scipy matplotlib tensorflow keras_tuner pytorch sklearn imblearn   

How to Use

1. Open DIabetes Neural Network.ipynb in Jupyter Notebook or JupyterLab.
2. Run all cells in order to reproduce the full pipeline.
3. Modify the Parameters cell to experiment with both your neural network architecture (e.g. number of layers, units per layer, activation functions, dropout rates) and your preprocessing pipeline (e.g. SMOTE ratio, number of PCA components, LDA dimensions, polynomial feature degree, etc.).

---

Contributing

Contributions are welcome! Please open issues or submit pull requests at
https://github.com/pablo-reyes8

License

This project is licensed under the Apache License 2.0.