This study compares the quality of synthetic data generated by four generative AI algorithms: Conditional Generative Adversarial Networks (CGAN), Variational Autoencoders (VAE), Synthetic Minority Over-sampling Technique for Regression (SMOTER), and Time Generative Adversarial Networks (TimeGAN). These models are trained on groundwater well data sourced from the Hanford site in Washington, USA, which has substantial missing data samples. Using these models, additional data were generated to replace these missing values. The datasets generated from each model are evaluated using a Gradient Boosting algorithm, and their data distributions are visualized. Comparative analysis reveals that the CGAN model produced synthetic data with a probability distribution closest to the original dataset. Furthermore, when the CGAN-generated dataset was combined with the original dataset, it yielded the best performance metrics among all generated datasets, demonstrating its excellence in effectively generating high-quality synthetic data. This highlights the potential of CGANs in addressing the challenges posed by missing data in environmental studies.