Conditional Data Synthesis Augmentation (CoDSA)

This repository contains the implementation for the paper "Conditional Data Synthesis Augmentation" (arXiv:2504.07426). CoDSA is a novel data augmentation framework that strategically generates synthetic data conditionally to improve machine learning models.

Repository Structure

Simulation Experiments

The Simulation/ directory contains code for:

• Section 4.1: Classification tasks
• Section 4.2: Regression tasks

Real-world Applications

The Real examples/ directory contains code for:

• Section 4.3: Anomaly detection
• Section 4.4: Age prediction
• Section 4.5: Image generation

Running Experiments:

• Each experiment folder contains detailed instructions in its README
• Jupyter notebooks are provided for step-by-step execution and visualization
• Example notebooks demonstrate the complete pipeline from data preparation to result analysis

Documentation

• Comprehensive implementation guides in each experiment directory
• Step-by-step workflow documentation and analysis reports in Jupyter notebooks