TSDCG: Tabular Synthetic Data with Code Generation LLMs

In the past years, there has been significant research on generating synthetic data, especially using Generative AI. Data engineering and analysis play crucial roles for any business in the study of past trends and forecasting future trends. However, real-world data often contains biases, missing values, and duplications, which limits its effectiveness in variety of tasks. Such issues can lead to model overfitting, underfitting, and poor generalization while training. To address these challenges, there has been extensive research on various GAN and LLM architectures and their ability to generate high quality synthetic data. This paper introduces and investigates a novel approach of using code-generating LLMs and advanced data profiling, in generating high quality synthetic data to enhance the speed and quality of the data generated.