A Data Preprocessor for Bayesian Model Averaging-Based Synthetic Dataset Detection Assessment Framework

AI generated synthetic data is increasingly prevalent, yet the ability to determine whether a dataset is synthetic remains limited. Generic AI detectors can identify synthetic content in some datasets, but their accuracy varies considerably. Specialized detectors perform well only on the specific types of datasets for which they were trained. Although many algorithms exist for detecting synthetic data, no generalized framework currently allows multiple detection methods to be combined to produce an overall probability that a dataset contains synthetic content. This study introduces a tool designed for broad organizational use that implements a generalized framework based on Bayesian Model Averaging (BMA) to improve the probability of detecting synthetic data. The BMA Data Preprocessor uses Python and established BMA formulas to automate the calculation of a final detection probability using a structured input file. The study’s findings show that combining multiple detection methods through BMA improves the resulting probability compared with relying on a single detection method.