ZeoPropNet: A Machine Learning Framework for Zeolite Property Prediction

📌 Overview

ZeoPropNet is a machine learning framework designed for predicting various zeolite properties, particularly the framework density, using neural networks.This project was developed as part of a student internship at Laboratório Associado LSRE-LCM (Porto, Portugal) during July–August 2024.

Built with TensorFlow and Keras, it supports:

• Cross-validation for robust model evaluation
• Hyperparameter tuning with Keras Tuner
• Model training and evaluation with automated scripts
• Configuration-driven experimentation for flexible model adjustments

📖 You can check the full project report and documentation here.

The datasets used were sourced from my Web Scraping project, which compiled zeolite data from the International Zeolite Association (IZA). By extracting composite building units, I generated fingerprints as input for the neural network.

---

🚀 Key Features

• Cross-Validation: Ensures robust model evaluation.
• Hyperparameter Tuning: Uses Keras Tuner for model optimization.
• Model Training and Evaluation: Provides automated scripts for training, testing, and evaluating models.
• Configuration-Driven Approach: Enables easy modification of model architecture and training parameters through config files.

---

📁 Directory Structure

MatPropNet/
|├── config_files/       # Configuration files for different models and datasets
|├── data/              # Datasets used for training and testing
|├── hyperband/         # Results from hyperparameter tuning
|├── main/              # Scripts for training, testing, and evaluation
|   |├── data.py            # Functions for data loading and preprocessing
|   |├── main_cv.py         # Script for cross-validation
|   |├── main_hyper.py      # Script for hyperparameter tuning
|   |├── model.py           # Functions for building and training models
|   |├── saving.py          # Functions for saving model history
|   └── stats.py           # Utility functions for statistical analysis
|└── models/            # Saved models and associated metadata

---

⚡ Getting Started

1. Install Dependencies

Ensure TensorFlow, Keras, and other required libraries are installed.

2. Prepare Data

Select the property to predict and the corresponding dataset in data/. The datasets contain fingerprints generated from zeolite composite building units.

3. Configure Models

Modify the config files in config_files/ to adjust model architecture and training parameters.

4. Run Scripts

Execute scripts from main/ to train, evaluate, and fine-tune models.

---

💻 Example Usage

Run Cross-Validation:

python3 main/main_cv.py config_files/density_iza.cfg

Perform Hyperparameter Tuning:

python3 main/main_hyper.py config_files/density_iza.cfg

---

✅ Conclusion

This project provides a structured framework for zeolite property prediction using machine learning, enabling efficient model training and evaluation.

python3 main/main_hyper.py config_files/density_iza.cfg