📊 Synthetic Data Generator (Streamlit)

The Synthetic Data Generator is a tool for generating IID (Independent and Identically Distributed) and Non-IID datasets with customisable parameters. Built with Python and Streamlit, this app enables users to create structured synthetic data for machine learning, research, and data exploration.

---

🛠️ Features

• Generate IID Data: Features are drawn independently from a standard normal distribution.
• Generate Non-IID Data: Feature distributions are adjusted to depend on the target variable.
• Customisable Data Generation:
  • Select the number of samples.
  • Choose number of features and target classes.
  • Introduce class imbalance (optional).
  • Set a random seed for reproducibility.
• Automated Validation:
  • Chi-Square Test: Checks feature dependence on the target.
  • Mutual Information Analysis: Measures how much features reduce uncertainty about the target.
  • Feature Correlation Matrix: Evaluates relationships between generated features.
• Download Full Dataset: Export your generated dataset in CSV format.
• User-Friendly UI: Built with Streamlit for seamless interaction.

---

🧬 Project Structure

SyntheticDataGeneratorApp
├── .streamlit/
│   └── config.toml
├── assets/
│   └── gifs/
│       └── dataGeneratorApp.gif
├── streamlit_app/
│   ├── pages/
│   │   ├── 01_1️⃣_IID_Data.py
│   │   └── 02_2️⃣_Non_IID_Data.py
│   └── 00_ℹ️_Info.py
├── LICENSE                 
├── README.md               
└── requirements.txt   

🎯 Motivation

Finding synthetic data for machine learning experiments can be challenging, especially for research projects.
This tool was created after experiencing difficulties in finding high-quality synthetic datasets during my MSc project.
With this app, you can generate and download free synthetic data instantly, tailored to your needs.

---

🏗️ How It's Built

The Synthetic Data Generator leverages the following Python libraries:

• Streamlit – Interactive UI for dataset generation.
• NumPy – Random number generation and data structuring.
• Pandas – Dataframe handling and CSV export.
• Seaborn & Matplotlib – Data visualisation.
• SciPy – Chi-Square tests for statistical independence.
• Scikit-learn – Mutual Information analysis.

---

📊 Understanding IID vs. Non-IID Data

Type	Characteristics
IID (Independent and Identically Distributed)	Features are generated independently from a normal distribution. No feature is related to the target.
Non-IID (Non-Independent and Identically Distributed)	Feature distributions shift based on the target class (different mean & scale per class). This ensures dependence between features and target.

How We Validate IID vs. Non-IID Data

✅ Chi-Square Test – Ensures features in Non-IID datasets are dependent on the target.
✅ Mutual Information – Measures feature-target relationships.
✅ Feature Correlation – Examines relationships between generated features.

---

🖥️ Using the App

1. Navigate to the Homepage (ℹ️ Info) to understand the differences between IID & Non-IID.
2. Go to "📊 IID Data" to generate an independent dataset.
3. Go to "📊 Non-IID Data" to generate a dataset where features depend on the target.
4. Adjust Parameters:
   • Select sample size, features, and classes.
   • Optionally, introduce class imbalance.
   • Set a random seed for reproducibility.
5. Generate & Visualise Data:
   • View feature distributions, correlation heatmaps, and mutual information scores.
6. Download Full Dataset as CSV.

---

🎬 Demo

Synthetic Data Generator App

---

⚠️ Known Issues & Fixes

Issue	Cause	Solution
Theme not working in deployment	.streamlit/config.toml is ignored by Streamlit Cloud	Move config.toml to ~/.streamlit/config.toml or apply st.set_page_config() directly.
Only 5 rows downloaded	Streamlit’s default table download only exports preview rows	Added st.download_button() for full dataset export.
Page refresh when downloading CSV	Streamlit re-runs on interaction	Implemented st.session_state to persist generated datasets.

---

🚀 Future Enhancements

• More Data Distributions: Support for uniform, exponential, and categorical distributions.
• Custom Feature Relationships: Users define correlations between features.
• Time-Series Synthetic Data: Generate sequential synthetic data.

---

🤝 Contributions

🔹 Have an idea to improve the app? Feel free to fork the repo and submit a pull request!
🔹 If you encounter any issues, report them on the GitHub Issues page.

---