This repository contains the implementation used for the research project:

"Synthetic Data Generation for Machine Learning Models Oriented to the Dielectric Characterization of Liquids Using a Dielectric Resonator"

• Author: Javier Villacampa Porta
• Supervisors: Miguel Monteagudo Honrubia & Francisco Javier Herraiz Martínez
• Institution: ICAI – Universidad Pontificia Comillas

This study evaluates and compares TVAE and CTGAN models for synthetic data generation to enhance machine learning models for liquid characterization. Results demonstrate that TVAE outperforms CTGAN in terms of efficiency, data quality, and predictive accuracy, making it a superior choice for dielectric characterization tasks.

Keywords: Deep Learning, Dielectric Resonator Sensor, Synthetic Data Generation (SDG), Variational Autoencoder (VAE), Generative Adversarial Networks (GAN)

✅ Synthetic Data Generation: Leverages TVAE & CTGAN to generate high-quality synthetic data for dielectric characterization
✅ Machine Learning Pipeline: Implements SVM, SVR, PCA, and data preprocessing for predictive modeling
✅ Performance Benchmarking: Evaluates models using f1-score, RMSE, and accuracy
✅ Visualization Tools: Interactive seaborn, matplotlib, and Plotly visualizations
✅ Data Profiling & Optimization: Monitors memory usage and performance

📦 Modelo-Spectral
│-- 📁 data/                  # Raw & processed datasets
│-- 📁 notebooks/             # Jupyter Notebooks for analysis
│-- 📁 models/                # Trained models & synthetic data
│-- 📁 reports/               # Research findings & documentation
│-- 📁 src/                   # Main codebase
│   ├── data_processing.py    # Data import & preprocessing
│   ├── synthetic_generation.py  # TVAE & CTGAN model implementation
│   ├── ml_models.py          # SVM & PCA-based ML models
│   ├── visualization.py      # Data visualization functions
│   ├── evaluation.py         # Model performance evaluation
│-- README.md                 # Project documentation
│-- requirements.txt          # Dependencies
│-- LICENSE                   # License information

1️⃣ Clone the repository

git clone https://github.com/villacampaporta/Modelo-Spectral.git
cd Modelo-Spectral

2️⃣ Install dependencies

pip install -r requirements.txt

3️⃣ Run the analysis

python src/data_processing.py
python src/synthetic_generation.py
python src/ml_models.py
python src/evaluation.py

1️⃣ Data Preprocessing: Loads and processes dielectric characterization data

2️⃣ Synthetic Data Generation: Implements TVAE & CTGAN for generating high-quality synthetic datasets

3️⃣ Machine Learning Models: Trains and evaluates SVM, SVR, and PCA-based models

4️⃣ Performance Evaluation: Compares synthetic vs. real data effectiveness using standard metrics

5️⃣ Visualization & Reporting: Generates insights through seaborn, Plotly, and Matplotlib

🔹 TVAE outperformed CTGAN in data similarity and predictive model performance

🔹 Generated synthetic data improved classification accuracy in dielectric characterization tasks

🔹 Memory-efficient & scalable solution for real-world applications

Programming: Python (Pandas, NumPy, Matplotlib, Seaborn, Plotly)

Machine Learning: Scikit-learn (SVM, SVR, PCA), Synthetic Data Vault (SDV)

Deep Learning: TVAE, CTGAN

Performance Monitoring: Memory profiling, computational efficiency

🔹 Extend model evaluation to other dielectric resonator-based datasets

🔹 Integrate reinforcement learning for adaptive synthetic data generation

🔹 Optimize data synthesis using hybrid VAE-GAN architectures

This project is licensed under the MIT License - see the LICENSE file for details.

Special thanks to ICAI – Universidad Pontificia Comillas for supporting this research.

📧 Email: javier.villacampa.porta@gmail.com

🔗 LinkedIn: linkedin.com/in/javiervillacampa

🌍 GitHub: github.com/villacampaporta

🔥 If you find this project useful, give it a ⭐ and contribute!